## The Technology Blindspot # Big AI Is Becoming Facebook. Your Privilege Is the Product. [%5D(/in/jdavidmorris/) [## JD Morris](/in/jdavidmorris/) COO & CoFounder LexAxiom | Agentic AI for the Business of Law | Kazeon → EMC | Dell · VMware · Cisco | Texas A&M MLS Law | Columbia MBA | Berkeley MBA | GW MEng | Auburn Physics April 23, 2026 **The Technology Blind Spot** On the morning of February 9, 2026, OpenAI began testing advertisements in ChatGPT for users on its free and $8-per-month Go tiers. That same morning, in San Francisco, Mrinank Sharma worked his last day at Anthropic. He had led the Safeguards Research Team since its founding. Before heading into work, he posted his resignation letter on X. The line that traveled further than he intended: “The world is in peril.” Two days later, on February 11, The New York Times ran Zoë Hitzig’s guest essay titled I Left My Job at OpenAI. Putting Ads on ChatGPT Was the Last Straw. In it, she argued OpenAI was making the mistakes Facebook had made, and she had quit. She had been a researcher at OpenAI for two years, helping shape safety policies and pricing models. Her resignation took effect the day the ads began. Three frontier AI labs. Two same-day departures at two different companies. One essay that named the pattern. The resignations are not a Silicon Valley drama. They are a vendor diligence signal for any firm whose associates are using ChatGPT, Claude, or Google’s Gemini on client matters. Model Rule 1.6(c) requires attorneys to make reasonable efforts to protect client information. What the AI labs’ own safety leaders now say about their employers’ trajectory is part of the facts-and-circumstances inquiry that “reasonable efforts” demands. **OpenAI: The Departures That Put Ads in Context** Hitzig is the sixth senior OpenAI researcher to walk out with a public indictment since February 2024. The pattern has been widely reported, and the arc matters because it supplies the context for the ad launch she resigned over. William Saunders and Daniel Kokotajlo both refused OpenAI’s non-disparagement clause at departure. Saunders, a three-year member of the technical staff, later testified before the Senate Judiciary Subcommittee on Privacy, Technology and the Law on September 17, 2024, that he had resigned because he had “lost faith that by themselves they will make responsible decisions about AGI.” Kokotajlo left the governance team in April 2024; the refusal cost him roughly $2 million in vested equity, which he later described as approximately 85% of his family’s net worth. Vox’s Kelsey Piper obtained the departure documents; four independent employment lawyers called the clawback language “egregious and unusual.” OpenAI reversed the policy within weeks. Ilya Sutskever co-founded OpenAI and served as its chief scientist. He resigned May 14, 2024. In a deposition unsealed in November 2025 as part of Musk v. OpenAI, Sutskever testified that he had authored a fifty-two-page memo to the independent directors accusing Sam Altman of “a consistent pattern of lying.” Jan Leike co-led the Superalignment team with Sutskever and resigned three days after him. His public thread on X on May 17, 2024 produced the most-cited external indictment of OpenAI’s trajectory: “safety culture and processes have taken a backseat to shiny products.” OpenAI dissolved the Superalignment team within days. Miles Brundage led AGI Readiness for six years. He left October 23, 2024. His farewell statement: “Neither OpenAI nor any other frontier lab is ready, and the world is also not ready.” OpenAI dissolved the AGI Readiness team with him. Hitzig timed her essay to the ad launch because the essay is about what the ads make possible. ChatGPT has generated, in her words, “an archive of human candor that has no precedent.” Users disclose medical fears, legal strategy, and intimate family questions because they treat the chatbot as private. Ads change the commercial logic of that archive. **Anthropic: The Parallel** Researchers who left OpenAI over safety concerns founded Anthropic in 2021. The company markets itself as the safety-first frontier lab. It operates under a Public Benefit Corporation charter and holds a governance structure designed to resist commercial pressure. February 2026 tested the design. Mrinank Sharma led the Safeguards Research Team, which Anthropic had launched in early 2025 to build production defenses against AI-assisted bioterrorism and other catastrophic risks. He holds a machine learning doctorate from Oxford. His resignation letter, published on his Substack on February 15, describes the reason for leaving in restrained language. The passage that circulated most: “I’ve repeatedly seen how hard it is to truly let our values govern our actions. I’ve seen this within myself, within the organization, where we constantly face pressures to set aside what matters most.” Two other senior Anthropic researchers, Harsh Mehta and Behnam Neyshabur, announced their departures the same week. Fifteen days later, on February 24, 2026, Anthropic published Responsible Scaling Policy version 3.0. The Responsible Scaling Policy is the framework under which the company commits to pause training or deployment if safety mitigations cannot be made adequate. RSP v1.0, released in September 2023, contained hard pause commitments. RSP v3.0 does not. Anthropic describes the change in its own posted rationale. The rewrite, the company writes, “represents a change from our previous RSP, driven by a collective action problem.” The argument: if one frontier lab pauses unilaterally while others continue, the pause does not reduce aggregate risk. The new policy replaces operational pause commitments with a Frontier Safety Roadmap, quarterly Risk Reports, and a commitment to advocate for industry-wide standards. Anthropic’s own policy team acknowledged the shift from binding commitments to reporting. The primary source here is Anthropic’s own website. The company posted the rewrite with a defended rationale, a redline log against prior versions, and an open invitation to critique. That transparency matters. It contrasts directly with OpenAI, which dropped “safety” from its IRS Form 990 filings without public announcement until The New Yorker reported it in April 2026. Honest trade-offs disclosed in public are different from trade-offs made quietly. Anthropic is doing the former. The trajectory it discloses, however, is still the trajectory Leike described at OpenAI in 2024: safety commitments that held when the lab was small, softening when commercial stakes grew. **Google DeepMind: The Quieter Version** Google DeepMind departures get less press, which is partly a function of scale. The research division sits inside a public company with more than 180,000 employees. Individual resignations do not land on the front page. Nicholas Carlini’s did. Carlini spent seven years at Google Brain and then Google DeepMind after the Brain-DeepMind merger. He is one of the most cited researchers in adversarial machine learning. On March 5, 2025, he published a public resignation memo on his website explaining why he was moving to Anthropic. The memo names the concern plainly: DeepMind leadership, after the merger, no longer supported the security and privacy research he believed was essential. He writes that he fought for that support for several years before concluding the structure would not change. Carlini’s departure did not come with a “world is in peril” line. It came with a seven-thousand-word technical explanation of why the institutional processes had made his research harder to publish. That is a different kind of indictment. It is not less serious. It says the infrastructure for doing safety research had eroded from inside, not that the company abandoned the topic in public. Richard Ngo’s path is a different kind of signal. Ngo was a research scientist on DeepMind’s AGI safety team from 2020 to 2023, then moved to OpenAI as a governance researcher. He left OpenAI in November 2024 citing, in a message to colleagues, “a lot of unanswered questions about the mission.” He is now independent. A safety governance researcher who cannot find a frontier lab that supports his work is a data point that does not fit on any one company’s ledger. It sits on the industry’s. The structural picture is starker still. Mustafa Suleyman co-founded DeepMind in 2010. He left the lab under difficult circumstances in 2019 that had nothing to do with safety concerns. His subsequent trajectory, however, illustrates the gravitational field in which the industry operates: DeepMind to Google policy role, Google to the venture firm Greylock, Greylock to Inflection AI, Inflection to Microsoft, where he now runs Microsoft AI. Business Insider reported in January 2026 that at least eleven senior Google AI and cloud executives left in 2025, most to Microsoft. Institutional knowledge does not stay inside any one lab. It moves with the commercial gradient. **Three Labs, One Pattern** Different specifics. Same arc. At OpenAI, safety teams were built, the leaders quit, the teams dissolved. At Anthropic, the safety leader quit, and within two weeks the governing safety policy softened its binding commitments. At Google DeepMind, the institutional processes around safety research degraded until a senior researcher left publicly to join a different lab. Three different mechanisms. One common pressure: commercial imperatives scaled faster than the infrastructure supporting safety commitments. This is not an indictment of any single executive or any single lab. It is a structural observation about what happens when safety commitments made at the scrappy-startup stage meet the gravitational pull of enterprise revenue, IPO math, and sovereign-scale investment. Facebook did not set out to become the company the 2019 Federal Trade Commission consent order documented. It became that company by making a series of defensible local decisions, each one driven by the same commercial logic. Hitzig’s phrase is the one worth holding: “the most detailed record of private human thought ever assembled.” Three labs are building versions of that archive simultaneously. All three are under commercial pressure. At least one of the three has launched advertising. A second has softened its pause commitments. The third is a division of a public company whose fiduciary obligations run to shareholders. **The Morning of May 17** On the morning of May 17, 2024, Jan Leike opened X and typed a thread he knew would end working relationships with people he had respected for years. He had co-authored the Superalignment announcement in July 2023. He had helped negotiate OpenAI’s public commitment of 20% of its compute over four years to the problem of aligning systems “much smarter than us.” In the year since, the compute had not materialized. Experiments that should have taken weeks stretched to months. Research directions requiring large training runs became impractical. He posted the thread. OpenAI dissolved the Superalignment team within days. Leike joined Anthropic the same week. Twenty-one months later, Anthropic’s own safeguards lead resigned. This is the pattern legal technology vendors have watched from the outside. It is not a Silicon Valley story. It is a supply-chain story. **The Privilege Problem** On February 17, 2026, eight days after the OpenAI ads launched and eight days after Sharma left Anthropic, Judge Jed S. Rakoff of the Southern District of New York issued the written memorandum in United States v. Heppner. The ruling stripped attorney-client privilege from thirty-one documents the defendant had typed into Anthropic’s Claude chatbot. The reasoning turned on Anthropic’s written privacy policy: Anthropic collects inputs and outputs, uses them to train the model, and reserves the right to disclose data to third parties including “governmental regulatory authorities.” Judge Rakoff concluded that Heppner’s inputs were “not confidential” because the platform’s own policy disclosed the opposite. The ruling named Anthropic. The analysis applies to every major consumer AI platform. OpenAI’s Consumer Terms of Use state services are “AS IS,” disclaim all warranties including security, and authorize use of user content to “provide, maintain, develop, and improve” the services. Google’s consumer Gemini terms contain substantially similar language. All three frontier labs target their consumer terms at a market that is not legal work. Now overlay the 2026 events. OpenAI routes commercial advertisements through the archive its terms authorize the company to use. Anthropic softened its Responsible Scaling Policy two weeks after its safeguards lead resigned. Google DeepMind watched its most publicly vocal security researcher leave citing institutional barriers. A firm onboarding any of these tools in 2024 did so under one risk profile. The profile has changed. This is the Rule 1.6(c) analysis. The Rule requires reasonable efforts to prevent unauthorized disclosure. ABA Formal Opinion 477R (2017) frames “reasonable efforts” as a facts-and-circumstances inquiry that weighs sensitivity, cost, and likelihood of disclosure. A vendor whose data-handling posture has visibly shifted since the firm deployed the tool is a facts-and-circumstances input the analysis cannot ignore. [See Open AI Just Lost in Court: Judge Rakoff’s Ruling and What It Means for Attorney Privilege, The Technology Blind Spot (2026).] **The Objections** Three objections deserve direct engagement. Start with the disgruntled-employee argument. Critics point out that each of these researchers has context. Sutskever led a failed coup against Altman. Saunders and Kokotajlo share ties to the effective altruism movement. Sharma cited personal reasons and plans to study poetry. Carlini had a specific dispute with DeepMind leadership about his research area. The rebuttal is not that any single person is pristine. The rebuttal is that senior safety leaders at three independent frontier labs reached parallel institutional conclusions within a narrow time window. Coincidence does not scale across three companies with different cultures, different funding structures, and different governance models. Next, the collective-action argument. Anthropic makes it explicitly in the RSP v3.0 rationale: unilateral safety commitments do not reduce aggregate risk if competitors ignore them. The argument has force. If all frontier labs are on similar commercial trajectories, an individual firm pausing unilaterally only shifts market share. The argument is also the concession. It acknowledges that commercial pressure is the active variable and voluntary safety commitments bend under it. That is precisely the conclusion the outside observer, and the managing partner buying the product, needs to hold in mind. Finally, the enterprise-versus-consumer argument. Enterprise contracts from OpenAI, Anthropic, and Google offer stronger data-handling commitments than consumer terms. Zero-retention clauses are available. Subprocessor disclosure is available. This is true and important. It is also not what an associate is using when she pastes a deposition excerpt into the free tier of ChatGPT at eleven at night. The enterprise posture matters only to the extent the firm has enforced it, documented it, and trained its associates to use it exclusively. **The Thursday Protocol** The managing partner reading this does not need to decide whether OpenAI is becoming Facebook. She needs to make three calls by Thursday. First, a call to the firm’s IT vendor or managed services provider. The question is narrow: which AI tools, or AI features embedded in existing software such as Microsoft Copilot or the case management system’s summarization function, route prompts through OpenAI, Anthropic, or Google infrastructure. For each vendor identified, the follow-up is whether the current contract prohibits use of firm data for advertising, training, or commercial purposes the vendor has not yet disclosed. If the answer is not in writing, the tool is not fit for client matters. Next, a call to whoever drafts the firm’s engagement letter. That may be the managing partner herself. Add a clause disclosing which AI tools the firm uses, identifying the subprocessors, and flagging material changes to vendor data practices. ABA Formal Opinion 512 (2024) frames the obligation. The Opinion does not mandate a specific clause. It requires informed client consent when confidentiality concerns exist. The vendor landscape that existed when the firm adopted these tools is not the landscape that exists today. [See 17 Subprocessors Deep: A Due Diligence Framework for Legal Technology Procurement, The Technology Blind Spot (2026).] Last, a call to the firm’s cyber insurance broker. Ransomware coverage is standard. Vendor-change-of-use coverage is not. Confirm whether current coverage responds if a major AI vendor materially changes its data-handling terms, advertising posture, or safety policy after the firm deployed the tool. If the coverage does not respond, the firm is carrying a risk it has not priced. **The Archive** Zoë Hitzig timed her resignation to the moment the OpenAI product changed character. Mrinank Sharma walked out of Anthropic the same day. Nicholas Carlini had left Google DeepMind eleven months earlier. The firms that adopted these tools in 2024, under earlier terms and earlier safety postures, are still using them. The prompts your associates typed into ChatGPT, Claude, and Gemini this morning are not gone. They are in the archive. What these companies do with the archive has already begun to change. The people best positioned to tell you are the ones who just walked out the door. *This blog provides general information for educational purposes only and does not constitute legal advice. Consult qualified counsel for advice on specific situations.* **About the Author** JD Morris is Co-Founder and COO of LexAxiom, an Agentic AI platform for the business of law. Over a 25-year career, he has built and scaled enterprise technology products across Dell, EMC, VMware, and Cisco, including the first exabyte eDiscovery platform. He holds dual MBAs from Columbia Business School (Finance) and UC Berkeley Haas (Marketing), a Master of Legal Studies in Cybersecurity Law from Texas A&M, and a Master of Engineering from George Washington University. He writes The Technology Blind Spot on the intersection of emerging technology and law. Connect with him on LinkedIn at http://www.linkedin.com/in/jdavidmorris, on X at @JDMorris_LTech, or on Bluesky at @JDMorris-ltech.bsky.social.

Originally published on LinkedIn Newsletter: The Technology Blind Spot