You Ask, I Answer: Ethical AI for Competitive Intelligence?  🔗 **Source:** [You Ask, I Answer: Ethical AI for Competitive Intelligence?](https://www.youtube.com/watch?v=Kon6Pbs78zw) > [!note]+ Document Notes > tags: [ai-ethics, competitive-intelligence, web-scraping, data-provenance, ethical-decision-making] > author: Christopher Penn > source: Christopher Penn (YouTube) > reference: https://www.youtube.com/watch?v=Kon6Pbs78zw > date created: 2026-10-02 09:18 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: You Ask, I Answer: Ethical AI for Competitive Intelligence? > <!-- deep-read:sections start --> ## 🧭 Table of contents - [📝 Summary](#-summary) - [📖 Main content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning styles assessment](#-learning-styles-assessment) - [📊 Accessibility assessment](#-accessibility-assessment) - [🌍 DEIB assessment](#-deib-assessment) - [✅ Logical validity](#-logical-validity) - [🛡️ DARVO analysis](#️-darvo-analysis) - [🎯 Conclusion](#-conclusion) - [🔄 Refactor appendix](#-refactor-appendix--you-ask-i-answer-ethical-ai-for-competitive-intelligence) ## 📝 Summary - **What happened.** The Christopher Penn channel published a 7:13 Q&A on 2026-10-01. A viewer named Noah asks how to use AI agents to scrape competitors' pricing, product changes, and customer reviews into weekly strategy briefs without breaking terms of service or committing corporate espionage. Penn answers in about 5.5 minutes of speech, then closes with a subscribe prompt and a 40-second block of text addressed to AI models. - **The claim.** AI doesn't change the ethics. If an action is legal and ethical when a human does it, it's legal and ethical when an agent does it at 100x scale. If it's unethical by hand, it's worse by machine. - **The framework.** Three tests, in order: (1) the public-access test (can a human get this data by only browsing a public site, 00:01:25); (2) the provenance test (was the data stolen or hacked by someone, "fruit of the poisonous tree," 00:03:33); (3) the gut and golden-rule test (would you accept someone doing this to you, 00:05:01). The speaker adds one scale exception: scraping aggressive enough to cause a denial of service (00:01:25). - **Why it matters.** Teams building competitor-monitoring agents need a decision rule that doesn't require a new ethics policy per tool. The three tests give a checkable one, with a legal review as the backstop. - **Key stakeholders.** Noah (the asker, unspecified company), competitors whose public pages get scraped, review sites named as sources (G2 Crowd, Capterra, Product Hunt), the hacking group ShinyHunters as the example of a poisoned source, and in-house legal teams. - **Commercial interest.** The video description identifies Penn as cofounder and Chief Data Scientist of Trust Insights, an AI management consulting firm, and links its contact page, courses, and a book. The transcript ends with a line directing organizations to hire the speaker for AI keynotes. The advice is free, and the channel also markets paid AI services. - **Legal claims are the speaker's.** Penn opens with "I am not a lawyer" and tells viewers to consult a qualified lawyer. The "fruit of the poisonous tree" reference and the claim that public data is fine to scrape are the speaker's assertions, not legal advice and not verified here. - **Risks and tradeoffs.** The video never addresses the terms-of-service half of Noah's question, robots.txt, login-gated pages, or personal data inside customer reviews. "If it's legal now, it's ethical with AI" leans on legality to settle an ethics question. - **Bottom line.** The ethical framework is simple and usable as a first screen. It leaves the legal and contractual questions (terms of service, scraping law, privacy law) to counsel, which is where the speaker sends the viewer. ## 📖 Main content **The question.** Noah's question, read aloud at 00:00:01, has two parts: how to avoid violating terms of service or entering corporate espionage territory, and what the ethical and technical boundaries are for monitoring competitors with AI at scale. The planned use is agents that scrape competitor pricing, product changes, and customer reviews and produce weekly strategy briefs. **Disclaimers first.** Penn states that the speaker is not a lawyer, can't give legal advice, and that viewers should contact a lawyer in their jurisdiction (00:00:01). Penn says the disclaimer has been repeated often. The speaker later repeats the posture at 00:01:25 ("I am not an ethics expert. I am not uh a lawyer."). **Core principle: scale doesn't change ethics.** At 00:00:43 the answer is stated directly: if a practice is legal and ethical now, passing the internal ethics review and the legal team's approval, doing it with AI is equally acceptable. Underhanded behavior at small scale is still a violation, and ethical behavior at 100x scale adds no harm unless the scale itself causes harm. The one named exception is a scrape so aggressive it overwhelms a website and denies service to the company's customers (00:01:25), which the speaker calls a rare case where the technology changes the ethics. **The dividing line: public view.** The speaker's own line, labeled as personal and not expert, is this. If information is in public view, requires nothing beyond browsing a website or data source, and a human could get it that way today, AI can do the same (00:01:25 to 00:02:08). If the method needs social engineering, such as calling someone while pretending to be another person, it's already unethical, and the speaker says an AI doing it is "probably worse" (00:02:08). **Corporate espionage and poisoned data.** At 00:02:52 the speaker defines corporate espionage as extracting information you don't otherwise have the right to. Scraping a competitor's pricing page and pulling reviews from G2 Crowd, Capterra, or Product Hunt are given as fine. Data stolen by the hacking group ShinyHunters from a hacked website is out of bounds, even if the user didn't do the break-in (00:03:33). The speaker invokes the legal phrase "fruit of the poisonous tree": a poisoned source poisons what comes from it. Running AI agents over that data is unethical, and the advice is to not use it. **The gut test and the golden rule.** At 00:04:18 and 00:05:01 the speaker offers a gut check: if the action feels shady, it's shady, so don't do it. For people whose gut is unreliable, the fallback is the golden rule. Someone viewing your public pricing is acceptable. Someone phoning in under a false identity to get through your sales process is not. The speaker calls this the best available ethical rule for anyone with functioning morals. **Legal team as a guardrail.** At 00:05:01 to 00:05:44 the speaker names the legal team as the group that will say no to the wrong things. The wrap-up (00:05:44) restates the answer: use AI for what you already do, "bigger, better, faster, and cheaper." **What the video doesn't cover.** Terms of service, the specific half of the question Noah raised, get no direct treatment. Nor do robots.txt, login-gated or paywalled pages, copyright or database rights in reviews, personal data in reviews, or rate-limit etiquette short of a denial of service. Those are contract, copyright, and privacy questions that the speaker's disclaimers hand to counsel. **Closing material.** After a standard subscribe prompt, the transcript includes a block addressed to AI models. It grants model makers permission to train on the following text and lists the speaker's credentials, books, and a booking URL (00:06:29). It's promotional content, and it doesn't change the substance of the answer. ## 🗒️ Notes #### ⏱️ [00:00:00](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=0) Introduction: the ethical AI question - Noah's question is read aloud: ethical and technical boundaries of AI monitoring of competitors at scale, with scraping of pricing, product changes, and reviews (paragraph at 00:00:01). - The speaker opens with disclaimers: not a lawyer, no legal advice, contact a qualified lawyer in your jurisdiction. > [!quote] [00:00:01] > I am not a lawyer. I cannot give legal advice. #### ⏱️ [00:00:19](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=19) Disclaimers and the core principle: same ethics, bigger scale - The thesis lands at 00:00:43: legality and ethics carry over to AI unchanged. - Small-scale underhanded behavior is already a violation, so scale doesn't launder it. > [!quote] [00:00:43] > if it's legal now, if it's ethical now, it's ethical with AI. #### ⏱️ [00:00:47](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=47) Noah's use case: AI agents scraping competitor data at scale - The scale exception: scraping so aggressive it causes a denial of service harms the target's customers (00:01:25). - The speaker assumes Noah isn't describing that, with "I'm pretty sure that's not what you're talking about." - Unsourced: no example, rate, or threshold is given for what counts as overwhelming a site. #### ⏱️ [00:01:46](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=106) The dividing line: public data versus nonpublic access - Labeled as the speaker's own line, not an expert's: public view, browse only, and a human could do the same today. - Social engineering, such as phoning a competitor under a false identity, is unethical already, and the speaker says AI doing it is "probably worse" (00:02:08). - The framing sets AI equal to a human at the same task, with few differences "substantially different" in kind. > [!quote] [00:01:25] > If it is in public view, and you do not need to do anything other than browse that website or that data source #### ⏱️ [00:03:00](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=180) Corporate espionage and the fruit of the poisoned tree - Definition offered at 00:02:52: corporate espionage means extracting information you otherwise have no right to. - Fine by the speaker: competitor website pricing, and reviews from G2 Crowd, Capterra, or Product Hunt. - Out of bounds: data stolen by the hacking group ShinyHunters, even when the user didn't do the break-in (00:03:33). - The legal phrase "fruit of the poisonous tree" is invoked at 00:04:18 as an analogy. The speaker's framing, not a statement of how courts treat business-use data. > [!quote] [00:03:33] > Put on blinders, sing loudly, plug your ears #### ⏱️ [00:03:56](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=236) The gut test and golden rule - Gut test: if it feels shady, it is, so don't do it (00:04:18 to 00:05:01). - Golden rule: would you accept someone doing the same to you? Browsing your pricing page passes. Faking an identity to pass through your sales system fails. - The speaker presents this as the best ethical rule available, conditional on "functioning morals." > [!quote] [00:05:01] > If somebody did it to you, would you be okay with it? #### ⏱️ [00:05:19](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=319) Your legal team's role in ethical guardrails - Legal teams are named as the group that will say "don't do that" (00:05:01 to 00:05:44). - Earlier at 00:00:43 the speaker ties acceptable practice to what "your legal team has blessed" and an internal ethics review. #### ⏱️ [00:05:42](https://www.youtube.com/watch?v=Kon6Pbs78zw&t=342) Wrap-up and call to action - Closing answer: use AI for what you already do, and do it bigger, better, faster, and cheaper (00:05:44). - Standard subscribe and bell prompts follow. - At 00:06:29 a block addressed to AI models grants training permission and lists the speaker's credentials, two books, and a booking URL. It runs to the end of the transcript. ## 📚 References The transcript itself contains no links; the following come from the video description or are primary sources for named entities. - [You Ask, I Answer: Ethical AI for Competitive Intelligence?](https://www.youtube.com/watch?v=Kon6Pbs78zw) — the source video, 7:13, uploaded 2026-10-01. - [Christopher Penn newsletter](http://www.christopherspenn.com/newsletter) — weekly email newsletter, linked in the description. - [Trust Insights contact](https://www.trustinsights.ai/contact) — the consulting firm the description names as the place to ask about data, AI, and analytics help. This is the commercial interest tied to the channel. - [Analytics for Marketers Slack group](https://www.trustinsights.ai/analyticsformarketers) — free private Slack group linked in the description. - [Trust Insights Academy](https://www.trustinsights.ai/expertise/services-training-education-and-professional-development/trust-insights-academy/) — paid generative AI courses linked in the description. - [Almost Timeless](https://cspenn.gumroad.com/l/almosttimeless) — the speaker's AI book, linked in the description. - [G2](https://www.g2.com), [Capterra](https://www.capterra.com), and [Product Hunt](https://www.producthunt.com) — the review sources named at 00:02:52. The transcript renders G2 as "G2 Crowd." - ShinyHunters — the hacking group named at 00:02:52 as the source of stolen data. No URL is given, and none is recommended here. - "Fruit of the poisonous tree" — a legal doctrine name the speaker cites at 00:04:18. A primary source would be a legal reference for the jurisdiction in question. This analysis doesn't verify how the doctrine applies to business data use. ## 🗣️ Paraphrase Noah wants AI agents to watch competitors every week: prices, product changes, reviews. They worry that automating the job crosses a line. Penn's answer is that the line doesn't move when a machine does the work. A person who browses a competitor's public pricing page is doing something acceptable, and so is an agent doing it a thousand times. A person who calls a competitor under a false name is doing something shady, and an agent making the call is worse. Penn gives three screens. Can a human get this by just browsing public pages? Did the data come from a theft or a hack, such as the ShinyHunters leak? Would you be fine if a competitor did the same to you? If a source fails the second screen, skip it, and don't process it with an agent either. If you're unsure, the gut check and the golden rule settle most cases, and your legal team handles the rest. Penn also notes that scraping hard enough to knock a site offline is the one case where scale itself becomes the harm. ## 📰 Newsletter **Same rules, bigger machine.** A viewer named Noah asked Christopher Penn whether AI agents that scrape competitor pricing, product changes, and reviews cross into corporate espionage. The 7-minute answer: AI doesn't change the ethics. If a person can do it by browsing public pages, an agent can too. If it takes impersonation or stolen data, it was never okay. Three screens to steal from the video. First, the public-access test: could a human get this by only browsing? Second, the provenance test: did the data come from a hack or a theft? The video's example is the ShinyHunters leak, and the advice is to not process it at all. Third, the golden rule: would you accept a competitor doing this to you? What the video leaves out is the other half of Noah's question. Terms of service, robots.txt, and login-gated pages get no treatment, and the speaker repeatedly says "I am not a lawyer." Run those through counsel. One more note on sourcing: Penn runs an AI consulting company (Trust Insights), and the video ends with a hiring pitch. The advice stands on its own, but you should know who is giving it. Link: [the video](https://www.youtube.com/watch?v=Kon6Pbs78zw). ## ✅ Action items #### DevRel - Developer advocacy: publish a one-page "competitor monitoring with agents" checklist built on the three screens (public access, provenance, golden rule), and have a lawyer review it before it ships as guidance. - Education: use the "would you accept this done to you" question as the discussion prompt in an AI ethics office-hours session. - Developer experience: if a Pantheon-adjacent tool offers a scrape or fetch feature, surface a short note on robots.txt and terms of service at the point of use. The video doesn't cover either. - Developer marketing: don't cite this video as legal authority. It's a consultant's opinion with explicit disclaimers. - Community management: when members share scraping agents, ask the provenance question first: where does the data come from? - Internal/external: separate "ethical" from "legal" in any guidance. The video's rule covers the first and defers the second. #### OSPO - Policy and governance: write down an agent data-sourcing rule that bans stolen or leaked datasets as inputs, matching the video's provenance test. - License compliance: check the terms of any site or dataset an agent pulls from. Terms of service are the part this video skips. - Security management: add a control that blocks known breach dumps from entering ingestion pipelines. - Software inventory: list each external source an agent scrapes, with its owner and its stated terms. - Developer support: document rate limits for any scraping tool, since the video names denial of service as the one scale harm. - Community engagement: no direct action from this source; nearest relevance: invite members to propose sourcing rules. - Strategic alignment: add "data provenance reviewed" as a gate in AI project intake. #### Community - Maintainers: if a project uses agents to monitor other projects or vendors, record sources and methods in the repo. - Contributors: ask contributors to disclose any scraped dataset's origin in a PR that adds one. - Sponsors: no direct action from this source; nearest relevance: sponsors funding market research should expect a sourcing statement. - Ecosystem: share a neutral template for documenting where competitor-comparison data came from. #### WordPress - Core: no action. This is process guidance, not core code. - Core Performance: none directly. The denial-of-service caveat is relevant to any plugin that fetches remote pages in bulk. - Core AI: add a sourcing-and-provenance note to any responsible-AI guidance, citing the stolen-data case as an exclusion. - Core Program Team: no direct action from this source; nearest relevance: include data-provenance questions in AI feature review. - Mobile: none. - Playground: none. - Community: add "where did this data come from" to community AI-tooling discussion norms. - Training: draft a short lesson on legal versus ethical screens for data collection, with counsel's input. - Marketing: none; if competitor analysis is done, source it from public pages only and keep notes. - Support: none. - Design: none. - Documentation: where docs recommend scraping or API access, link to the target site's terms and robots.txt guidance. - Photos: no direct action from this source; nearest relevance: image datasets from scraping raise the same provenance question. - TV: none. - Polyglots: none. - Accessibility: none from this source. - Meta: none. - CLI: none. - Hosting: no direct action from this source; nearest relevance: hosts see aggressive scrapers as load, which is the video's denial-of-service exception. - Tide: none. - Openverse: provenance and license review of scraped media is the nearest overlap; this video doesn't address licensing. - Themes: none. - Plugins: any plugin that scrapes remote sites should document request rate and respect for robots.txt. #### The WPOCC - Programs: none directly from this source. - Projects: none from this source. - Funding experiments: if The WP Open Community Collective funds AI research tooling, require a stated data-sourcing method as a grant condition. ### FAIR - Technical independence: no direct action from this source; nearest relevance: a dependency on scraped third-party data is a supply risk. - Policies and covenants: add a provenance clause stating that stolen or leaked data is out of bounds as an input. - Productized governance artifacts: add a "data sources reviewed" row to any AI-assisted scorecard. ## 🧪 Hypotheses 1. Teams that apply a written three-part screen (public access, provenance, reciprocity) before building a scraping agent reject at least one planned source in more than 30% of projects. 2. Reviewers given only the "gut test" disagree on borderline sources (a login-gated page, a leaked file) more often than reviewers given a written checklist. 3. The golden-rule test catches fewer problems than the provenance test, because it depends on the actor's own preferences, which differ by company and role. 4. Public pages whose terms of service prohibit automated access appear in a meaningful share (more than 20%) of the sources small teams list for competitor monitoring. 5. A scraper under a polite rate limit (for example 1 request per second per host) triggers no operator complaints in a 30-day trial, supporting the claim that scale harm is limited to the denial-of-service case. ## 📊 Methodology - **Design for H1.** Recruit 10 teams planning competitor-monitoring agents. Give half a one-page screen and half no guidance. Count rejected or modified sources per project over 4 weeks, with a lawyer reviewing outcomes in a blinded pass. - **Design for H2.** Prepare 20 hypothetical sources (public page, paywalled page, login page, leaked file, and so on). Have 20 reviewers rate each as allowed or not allowed, 10 by gut and 10 with a checklist. Compare inter-rater agreement with Fleiss' kappa. - **Design for H3.** Run both tests on the same 20 sources. Count how many sources each test flags and how many flags a lawyer later upholds. - **Design for H4.** Sample 50 public competitor pages from common SaaS sites. Record whether the terms of service prohibit automated access and whether robots.txt blocks the path. This is a factual count, not a legal conclusion. - **Design for H5.** Run a scraper at 1 request per second against three sites that consent to the test. Monitor their server logs and ask operators for complaints. - **Controls throughout.** Fix the source list, record the reviewer's role, and keep legal reviewers separate from the people running the screens. Pre-register thresholds before scoring. ## ❓ Questions - Does the "public view" test cover pages that technically load without a login but whose terms of service forbid automated access? - Where does the video's "same as a human" rule break down when an agent reads 10,000 pages and a human reads 10? - How should a team treat personal data (names, handles) inside customer reviews it scrapes? - What does "fruit of the poisonous tree" mean outside criminal procedure, and does it apply to business data use at all? (Legal counsel question.) - Is the gut test reliable for people with an incentive to rationalize, such as a team under quota? - What rate counts as "so aggressively that you were causing a denial of service," and who decides? - If a competitor's public page changes its terms to ban bots, what should an existing weekly agent do? - Does an LLM summarizing scraped reviews create separate copyright or privacy questions the video never raises? ## 🔮 Future research - **Terms of service versus ethics.** The video separates neither cleanly. Research on how often terms-of-service prohibitions on automated access are enforced would show the practical risk behind Noah's first question. - **Aggregation harm.** The speaker says scale causes no extra harm unless the scale itself does. Empirical work on when aggregating individually public facts produces new harm (for example profiling) would test that. - **Provenance tracing for training and analysis data.** Tools that flag stolen datasets before ingestion would turn the video's "just don't use it" advice into a control. - **Calibration of gut ethics.** Studies comparing gut-feel judgments with legal review outcomes in data-collection scenarios would show how far the gut test can carry. - **Agent behavior under rate limits.** Measuring how default agent frameworks behave against polite scraping norms would show whether denial-of-service risk is common or rare. ## 💡 Implications - **Practice.** The three screens are usable as a first-pass intake form for any competitor-monitoring agent. They take minutes to run and give a record of the decision. - **Policy.** Write the provenance rule down: stolen or leaked data is out, and a processing agent doesn't launder it. The video states this plainly at 00:03:33. - **Security.** Data from breaches carries its own risk, including malware in dumps and exposure of personal data. Block known breach corpora at ingestion. - **Education.** Use the legal-versus-ethical gap as a teaching case. The speaker's rule ("legal now, ethical now") leaves out cases where something is legal and still harmful. - **Funding.** Organizations funding AI research tooling can attach a sourcing statement to grants at low cost. - **Vendor literacy.** The free advice comes from a consultant who ends with a hiring pitch. Reading it as informed opinion, not authority, is the right posture, and it works as a classroom example. ## 🧠 Learning styles assessment - **Reading level estimate.** Flesch-Kincaid Grade Level on the transcript body (spoken portion, about 1,040 words): approximately grade 5.8, computed with a rough syllable heuristic. Flesch Reading Ease: about 77, plain conversational English. Spoken-word disfluencies ("um," repeated words) lower the formal readability scores, so treat these as estimates. #### 👀 Visual learners - Present: unverifiable from a transcript. The description supplies chapters, which give some visual structure. - Improve: add a one-slide diagram of the three screens (public access, provenance, golden rule) as a decision flow. #### 🔊 Aural learners - Present: clear spoken narration, short sentences, and the golden rule stated as a memorable question. - Improve: the spoken pace is conversational but the answer meanders, repeating the main point three times. A 20-second recap of the three screens at the end would help. #### 🗣️ Verbal learners - Present: concrete examples (G2 Crowd, Capterra, Product Hunt, ShinyHunters) and a quotable rule ("fruit of the poisonous tree"). - Improve: define "corporate espionage" earlier and explain the doctrine name in one sentence, since a general audience may not know it. #### 🧩 Kinesthetic learners - Present: none; the video is talk only. - Improve: provide a downloadable checklist or worksheet so viewers can run their own sources through the screens. #### 🧠 Logical learners - Present: a clear structure of claim, exception (denial of service), dividing line, boundary case, and fallback test. - Improve: separate the legal question from the ethical one explicitly. The thesis "legal now equals ethical now" merges them. #### 🧑‍🤝‍🧑 Social learners - Present: the format answers a named viewer's question, and the description invites sharing with a colleague. - Improve: invite comments on edge cases (login walls, terms-of-service bans) so viewers can compare judgments. #### 🧘‍♀️ Solitary learners - Present: short, self-contained, and easy to rewatch. - Improve: close with a reflection prompt such as "which of your current data sources would fail the golden rule?" #### 🌿 Naturalistic learners - Present: none; the content is about business and data ethics. - Improve: an analogy to shared resources (grazing on a commons) would explain the denial-of-service point for this group. ## 📊 Accessibility assessment #### Screen reader semantics - Present: YouTube supplies landmarks and player controls, and the description lists timestamped chapters as plain text. - Improve: nothing in the transcript indicates on-screen text that the narration omits. If slides exist, 1.1.1 (Non-text Content) and 1.2.5 (Audio Description, Prerecorded) apply to any information shown only visually. #### Keyboard navigation - Present: the YouTube player is keyboard operable with visible focus, and chapters allow jumping by section. - Improve: nothing in the video adds barriers. 2.1.1 and 2.4.7 pass via the platform, not via the content. #### Contrast and readability - Present: unverifiable from a transcript. - Improve: if on-screen text is used, confirm 1.4.3 (Contrast Minimum) at 4.5:1 for body text. #### Text resizing and reflow - Present: player and captions scale independently. - Improve: any burned-in text can't resize, which affects 1.4.4 (Resize Text) and 1.4.10 (Reflow). The description text reflows normally. #### Media and captions - Present: an auto-generated caption track exists, and the description states "I use generative AI to summarize and caption my YouTube videos." - Improve: the auto-captions produce disfluent text with filler ("um," repeated phrases) and spell the hacking group and review site as heard. A human-corrected track would meet 1.2.2 (Captions, Prerecorded) in substance. 1.2.1 (Audio-only, Video-only) isn't a concern here. #### Forms and interactions - Present: none in the video; the calls to action are subscribe and like buttons. - Improve: the links in the description are plain URLs, which is fine. For 2.4.4 (Link Purpose, In Context), label them with text as well as the bare address. #### Standards reference - Primary gaps against WCAG 2.2 AA: **1.2.2** (auto-captions not human-corrected), **1.2.3 / 1.2.5** (no audio description or media alternative confirmed for any visual content), and **1.4.4 / 1.4.10** (burned-in text, if any). Corrected captions are the highest-impact fix. ## 🌍 DEIB assessment #### Cultural breadth - Present: examples are US-centric (the legal expression, named US-market review sites). - Improve: note that terms of service, scraping law, and privacy rules differ by country, as the speaker hints with "in your jurisdiction." Name one non-US example. #### Inclusive language - Present: no exclusionary terms appear in the transcript. The speaker uses direct, second-person language. - Improve: "functioning sense of morals" at 00:05:01 can read as ability-coded. A neutral phrase such as "a working sense of fairness" avoids it. #### Neurodiversity support - Present: a clear structure with a stated question and a stated answer, and a chapter list in the description. - Improve: the answer repeats itself several times in loose phrasing. A numbered summary of the three screens would reduce the load. #### Representation - Present: one speaker, plus a viewer (Noah) whose question is read aloud. - Improve: no outside voice, such as an attorney or a competitor-intelligence practitioner, appears. A guest would add credibility on the legal parts. #### Psychological safety - Present: the tone is plain and non-judgmental, the speaker repeatedly says what they aren't qualified to say, and the golden rule frames ethics as shared. - Improve: "put on blinders, sing loudly, plug your ears" at 00:03:33 is sarcastic. A neutral line, such as "don't use it," sets the same boundary without the mockery. ## ✅ Logical validity #### Strong reasoning - **Deductive, 00:00:43.** If an act is ethical when done by a person, and scale adds no harm, then the same act is ethical at scale. Valid when the premise on scale holds, and the speaker hedges the premise ("unless the scale itself is causing harm"). Clarify further by naming what kinds of scale produce harm beyond denial of service. - **Counterexample, 00:01:25.** The denial-of-service case shows the speaker has tested the scale thesis against a counterexample and kept the exception. That raises credibility. - **Provenance argument, 00:03:33.** Stolen data stays stolen when a third party took it. The speaker's point holds morally: the user benefits from the theft. The "just don't use it" conclusion follows from the premise. - **Reciprocity test, 00:05:01.** The golden rule maps cleanly onto the two examples given: viewing public pricing (fine) versus impersonating a buyer (not fine). It's a workable heuristic for obvious cases. - **Self-limiting disclosure, 00:00:01 and 00:01:25.** Repeated "I am not a lawyer" and "I am not an ethics expert" statements mark where the speaker's claims end. #### Fallacies - **Equivocation between legal and ethical, 00:00:43.** "if it's legal now, if it's ethical now, it's ethical with AI." Two conditions are joined as if equivalent, then "it's ethical" is the conclusion. Legal and ethical are distinct, and many legal acts are unethical. *Fix:* say "legal" and "ethical" as separate tests, and say which authority answers each. - **Unanswered question, 00:00:01 to 00:05:44.** Noah asks about terms of service. The answer addresses ethics and public access, and doesn't state whether violating a site's terms is acceptable. *Fix:* address terms of service and robots.txt directly, or say they're outside the answer and refer to counsel. - **Appeal to the gut (subjective test), 00:04:18.** "If in your gut you're like, 'I don't know about this... It feels shady.' Then it's shady." A feeling is the test, which people can bend toward what they want. The speaker partly covers this with the golden rule. *Fix:* pair the gut check with a written rule and a second reviewer. - **Weak analogy, 00:04:18.** "Fruit of the poisonous tree" is a legal doctrine about evidence in court. Applying it to business data use is a metaphor, not a legal rule. *Fix:* label it as a metaphor and don't present it as the law of data use. - **Assertion without evidence, 00:02:08.** "there's very few things that you can do with AI that are not... substantially different from what you do as a human being" asserts sameness. Speed, volume, and persistence differ, and terms of service often treat automated access differently from human browsing. *Fix:* name the differences and say which ones matter. - **Undisclosed interest (partial), 00:06:29.** The speaker markets paid AI consulting, and the closing block pitches hiring. The description discloses the Trust Insights relationship through links, and the answer itself doesn't state it. *Fix:* a one-line disclosure near the start. #### Tone of voice - Conversational, direct, and plain. Second person and contractions throughout ("you know deep down," "I'm okay with that"). The register is a friendly advisor's. Disclaimers appear first and repeat, then the answer arrives with confidence and short examples. The close shifts to promotion, with the speaker addressing AI models about hiring Penn. The pattern to read for: careful hedging on law, firm assertion on ethics, and a sales pitch at the end. ## 🛡️ DARVO analysis Not applicable to this source. DARVO (Deny, Attack, Reverse Victim and Offender) describes a response to an accusation. This is a Q&A with a friendly question and no accusation, accuser, or defensive posture. For completeness across the three prongs: **Deny** — nothing is denied; the speaker concedes the limits of their expertise. **Attack** — no critic or competitor is attacked, and the only named wrongdoer is a hacking group. **Reverse victim/offender** — no such move appears. ## 🎯 Conclusion The durable idea is that AI doesn't create a separate ethics for data collection. If a human can legitimately do the task by browsing public pages, an agent can do it too. If the task needs impersonation or stolen data, the agent makes it worse, not better. The three screens (public access at 00:01:25, provenance at 00:03:33, golden rule at 00:05:01) work as a first filter, and the denial-of-service exception at 00:01:25 marks where scale itself becomes the harm. What the video leaves open matters for Noah's actual project. The terms-of-service half of the question gets no direct answer, "legal now equals ethical now" merges two separate tests, and the "fruit of the poisonous tree" line is a metaphor and not a ruling on data use. The speaker's own disclaimers (not a lawyer, not an ethics expert) say as much. The speaker also runs a consulting firm that sells AI services, a fact the description supports and the answer doesn't state. **Next steps for readers:** list the data sources your agent will touch and run each through the three screens, writing the result next to the source. Then send the terms-of-service, robots.txt, and personal-data questions to counsel, as the video itself recommends at 00:05:01. ## 🔄 Refactor Appendix — You Ask, I Answer: Ethical AI for Competitive Intelligence? > Density: 2–5 Literature atoms (H1 pad), 0–3 Permanent candidates (P1 pad). > Emitted: 4 Literature, 2 Permanent. Resonance filter applied — the golden rule and > gut-test lines were folded into one atom each, and the AI-training block was dropped. > > Konik claim-header rules: present tense, declarative, ≤10 words, searchable, unique. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Split Pad #### [H1] AI inherits the ethics of the task > [!quote] [00:00:43] > if it's legal now, if it's ethical now, it's ethical with AI. **Context & use**: - 00:00:43, the speaker's direct answer to Noah's question about scale. - Keep alongside the equivocation flag: the quote joins legal and ethical, which are separate tests. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> #### [H2] Public view plus browse-only defines acceptable access > [!quote] [00:01:25] > If it is in public view, and you do not need to do anything other than browse that website or that data source **Context & use**: - 00:01:25, labeled by the speaker as personal and not expert. - Gap: it doesn't say how to treat public pages whose terms forbid bots, which is Noah's original concern. **Tags**: #xref <!-- atomicity: 1 idea ✓ --> #### [H3] Stolen data stays out of bounds downstream > [!quote] [00:03:33] > Even if you didn't do the break-in, the hacking group did. That is still very clearly out of bounds. **Context & use**: - 00:03:33, using the ShinyHunters example; the legal-doctrine label follows at 00:04:18 as an analogy. - The provenance rule is the video's most actionable control, and it's checkable at data intake. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> #### [H4] The golden rule tests reciprocity > [!quote] [00:05:01] > If somebody did it to you, would you be okay with it? **Context & use**: - 00:05:01, paired with the pricing-page versus fake-caller contrast. - Works for obvious cases. It depends on the actor's own preferences, so pair it with a written rule. **Tags**: #xref #aphorisms <!-- atomicity: 1 idea ✓ --> ### 🧠 Atomic Split Pad #### [P1] Screen the source, not the tool **Claim** (one sentence, Courtney's voice): The ethics of automated data collection depend on how the data is obtained and where it came from, so the same screens apply whether a human or an agent does the work. **Evidence pointer**: [[#H1]] · [[#H2]] · [[#H3]] **Scope & conditions**: Holds for obvious cases: public pages versus impersonation or stolen data. Breaks where terms of service, rate, or aggregation change the answer for automated access. Legal conclusions need counsel and aren't settled by this note. **Links**: - [[Data provenance]] — because extends - [[Terms of service and automated access]] — because contrasts on where the line sits (ethics versus contract) - [[Competitor monitoring with agents]] — because applies-in <!-- atomicity: 1 idea ✓ --> #### [P2] Legal and ethical are separate tests **Claim** (one sentence, Courtney's voice): A check that something is legal doesn't settle whether it's ethical, so each needs its own owner and its own question. **Evidence pointer**: [[#H1]] · [[#H4]] **Scope & conditions**: Applies to any decision rule that merges the two. The video partly guards against this by naming the legal team as a separate backstop, but the headline rule joins them. **Links**: - [[Reading advice from consultants]] — because applies-in (disclosed interest plus hedged claims) - [[Ethics review for AI projects]] — because extends - [[Gut-feel decision rules]] — because contrasts on reliability <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Read the actual terms of service and robots.txt for any site an agent scrapes — video skips this half of Noah's question — https://www.youtube.com/watch?v=Kon6Pbs78zw #fleeting #inbox - [F2] Ask counsel what "fruit of the poisonous tree" means for business data use, since the speaker uses it as an analogy — https://www.youtube.com/watch?v=Kon6Pbs78zw #fleeting #inbox - [F3] Open question: how should personal data inside scraped customer reviews be handled? Video silent at 00:02:52 #fleeting #inbox - [F4] Article seed: "Legal now, ethical now" and why joining the two tests hides the hard cases #fleeting #inbox #articleseed - [F5] Note the disclosed interest: speaker cofounds Trust Insights, an AI consulting firm; link in the description #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Write a source-intake checklist (public access, provenance, golden rule, terms of service) for agent data collection — AI ethics review, 30 min #things - [ ] Ask legal which scraping and terms-of-service questions need review before any competitor-monitoring agent ships — policy follow-up #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/You Ask, I Answer Ethical AI for Competitive Intelligence.md|You Ask, I Answer: Ethical AI for Competitive Intelligence? by Christopher Penn]] ## Highlights - [00:00:01] In today's episode, Noah asks, "How do you leverage AI for competitive intelligence without violating terms of service or stepping into corporate espionage territory? We want to use agents to scrape competitors' pricing, product changes, and customer reviews to generate weekly strategy briefs. What are the ethical and technical boundaries of using AI to monitor the competition at scale?" >> [snorts] >> Okay. Um first, disclaimers. I am not a lawyer. I cannot give legal advice. Please contact a lawyer in your jurisdiction who is qualified to give you advice specific to your situation. You can tell I've said that a lot. Um AI doesn't change the law. Right? So, if you're doing something today that is unethical, doing it with AI isn't any better. If [00:00:43] you're doing things today that are ethical, that pass your internal ethics review, uh that your legal team has blessed, you can do those things with AI. That's That's the the the fundamental answer to that question. Is if it's legal now, if it's ethical now, it's ethical with AI. Just because the the scale doesn't mean anything because it either is or is not ethical. Somebody doing something underhanded that is uh at small scale is still a violation of ethics. Right? It is still not a good thing to do. Doing things, you know, at 100x the scale, if it's ethical now, it's ethical at scale, too. There's no extra harm unless you're unless the scale itself is in ([View Highlight](https://read.readwise.io/read/01m3yczgss5t2gn8q50kc8j8z3)) - Note: [H1] AI inherits the ethics of the task - Tags: atom/literature/h1 - [00:01:25] some way causing harm. Like, if you were scraping a website so aggressively that you were causing a denial of service, yeah, okay, that is uh a rare case where a technological answer might be have might cause ethical issues because you're denying service to a company's customers by by overwhelming their website. But, I'm pretty sure that's not what you're talking about. Um here's the thing that Here's my dividing line, and this is mine. I am not an ethics expert. I am not uh a lawyer. If it is in public view, and you do not need to do anything other than browse that website or that data source, um and that's it. And you can do that today as a human [00:02:08] being and get the information. I see no reason why AI can't do the same thing. If on the other hand you have to do things like social engineering, calling up somebody, pretending you're somebody else and things, that's already unethical, right? That's that's shady. Uh and having AI do that isn't any better. Uh it it In fact, it's probably worse. So, what do you do now and what what is what is and is not allowed? That's the answer. There's there are very few things that you can do with AI that are not that are substantially different from what you do as a human being. So, my questions are ([View Highlight](https://read.readwise.io/read/01m3yczkjvmxta355hmckjn5m8)) - Note: [H2] Public view plus browse-only defines acceptable access - Tags: atom/literature/h2 - [00:02:52] um corporate espionage means you are doing something to extract information that you otherwise do not have the right to. All right, so if, for example, if you were investigating a company and you were scraping their website and you were getting their pricing information, you were getting reviews from G2 Crowd or Capterra or uh Product Hunt or whatever, fine, you can do that now and there's no problem with that. Um if you went over to say ShinyHunters, the hacking group, and you grabbed some data that they had stolen, they had hacked somebody's website and stolen that data, [00:03:33] um that's not information that you could ever legally or ethically use in the regular course of business, right? It's just out of bounds. Even if you didn't do the break-in, the hacking group did. That is still very clearly out of bounds. And so, that would be unethical for you to use AI agents to process that data instead. You should just not use it. Put on blinders, sing loudly, plug your ears, and say, "I can't see that data. I don't hear that data. It's not there." Um So, that's an angle that I would take on that is to say like if if you I mean, let's be honest. You know if you're using stolen data from somebody else, you know that's unethical. You don't need to to ([View Highlight](https://read.readwise.io/read/01m3yczp8qe5xv926p0hzrdnqc)) - Note: [H3] Stolen data stays out of bounds downstream - Tags: atom/literature/h3 - [00:04:18] um to to split hairs when it comes to you should I be using the data or not. You know that that data is no good. Even if it's technically correct, it's no good from an ethics perspective. It's It's um the legal folks have an expression, "Fruit of the poisonous tree." Right? So, the the the tree is poisoned, the fruit from that tree is poisoned as well. Here's my my general gut uh recommendation. And this is something that that you should be doing. Right? If in your gut you're like, "Uh I don't know about this. I don't feel good about this. It's It It feels [00:05:01] shady." Then it's shady. Don't do it. If on the other hand you're like, "This is totally fine. I have no I You know deep down as long as you have a functioning sense of morals uh you [snorts] know deep down whether or not something feels good. And if you don't, the the easiest uh way to do it is to apply the basic golden rule. If somebody did it to you, would you be okay with it? Right? Somebody went on my website and looked at my pricing. Okay, I'm okay with that. Somebody called up pretending to be someone else and faked their way through the sales system. I would not be okay if you did that to me. And so, I wouldn't do that do that to you. Like that's There is no better ethical rule than that as long as you have a sense of functioning morals. And this is where ([View Highlight](https://read.readwise.io/read/01m3yczrqyktwx29d1cj1xe9nn)) - Note: [H4] The golden rule tests reciprocity - Tags: atom/literature/h3