OpenAI CTO Accuses Altman Of Lying Under Oath.

Mazen Al Sabbagh

— 8 min read
Share

TL;DR

  • Altman's Credibility Challenged: Former OpenAI CTO Mira Murati testified under oath that Sam Altman misled her on AI safety protocols and fostered a "chaos" and "distrust" culture among executives, delivering the most damaging insider account yet in Musk v. Altman. Separately, Shivon Zilis revealed Musk’s prior attempt to integrate OpenAI into Tesla, offering Altman a board seat, which complicates Musk's nonprofit mission claims. (Reuters, The Verge, CNBC)
  • EU AI Act Softened: Brussels reached a provisional agreement to simplify the AI Act, postponing high-risk system compliance to December 2027 and implementing the first notable rollback of European digital regulation under U.S. influence. The deal also explicitly bans AI nudification tools. (Reuters, Politico, EU Council)
  • OpenAI Enters Services Market: OpenAI's $10 billion joint venture is in advanced negotiations to acquire three AI services firms, aiming to internalize hundreds of engineers and consultants for enterprise AI deployments. This signals a strategic move by model providers into implementation, directly challenging traditional consulting giants. (PYMNTS, Reuters)
  • White House Reverses Course on AI Oversight: David Sacks departed his White House AI and crypto czar role as the administration prepares to implement AI model oversight measures he previously resisted. The White House is reportedly considering an Executive Order to review AI models akin to FDA drug approvals. (The Verge, The Hill, Bloomberg)
  • CoreWeave Earnings Today: CoreWeave reports its inaugural post-IPO quarterly earnings, with analysts projecting $1.97 billion in revenue and a loss of $0.89 per share. The results will offer critical insights into the valuation and sustainability of AI infrastructure investments, following Anthropic’s $200 billion commitment to Google Cloud. (Benzinga, Reuters)

Lead Story: OpenAI CTO Accuses Altman Of Lying Under Oath

Wednesday marked a significant inflection point in Musk v. Altman, with the most considerable damage to OpenAI's public narrative originating from its former leadership, not the plaintiff's counsel. The revelations reshape perceptions of the organization's foundational years.

Mira Murati, who served as OpenAI's CTO from 2018 to 2024 and briefly as interim CEO, provided sworn testimony that Sam Altman was dishonest regarding the safety evaluation process for new AI model releases. Murati reportedly stated she "couldn't trust Sam Altman's words" and accused him of cultivating an environment of "chaos" and "distrust" among top executives. She characterized the November 2023 board crisis, during which she assumed temporary leadership, as a period when "OpenAI was at catastrophic risk." Her testimony, delivered in a federal courtroom, followed a Met Gala appearance just days prior, underscoring the incongruous nature of the trial's unfolding drama. (Reuters, The Verge)

Concurrently, Shivon Zilis, a Neuralink executive and former OpenAI board member, offered testimony that introduced further complexity. Zilis, also a parent to four of Musk's children, described their arrangement as "platonic" and involving "just sperm," denying her role as an information conduit between Musk and OpenAI's board. However, WIRED previously reported on messages between Zilis and Tesla executives from 2017 detailing plans for a rival AI lab, potentially led by Altman or Demis Hassabis. Notably, CNBC and Bloomberg reported on Zilis's testimony that Musk had considered integrating OpenAI as a Tesla subsidiary and offered Altman a Tesla board seat in late 2017. This confirms that Musk, now litigating against OpenAI for deviating from its nonprofit mission, once sought to absorb it into a for-profit automotive enterprise. Emails presented in court further indicated Musk's concurrent efforts to recruit Altman to a "Tesla AI lab" while Altman was still on OpenAI's board. (CNBC, Bloomberg, WIRED)

Musk's subsequent endorsement of Murati on X, calling her "an amazing person," highlights how her testimony concerning Altman's candor aligns with his core legal claim. Business Insider's aggregation of multiple witness accounts paints a consistent picture of Altman's leadership style, depicted in terms more typically associated with challenging work environments than with the helm of a leading technology enterprise. Polymarket odds for Musk's victory remain at approximately 42%, with Altman expected to testify next week. (Business Insider)

In Other News

Brussels Blinks: EU Strikes Deal to Water Down Its Own AI Rules. After an unproductive trilogue session, EU negotiators finalized a provisional agreement that significantly eases the AI Act's implementation schedule. The compliance deadline for high-risk AI systems has been extended from August 2026 to December 2027, granting providers an additional 16 months. The deal also expands regulatory exemptions to small mid-caps, reduces specific documentation burdens, and strengthens the AI Office's enforcement capacity. Crucially, the agreement imposes an outright ban on AI "nudification" tools, making the creation of non-consensual sexual imagery via AI illegal under EU law. Politico characterized this as the "first significant delay of digital rules amid pressure from the U.S.," suggesting a capitulation to industry lobbying. This move bears watching as it sets a precedent for regulatory flexibility in the face of commercial interests. (Reuters, Politico)

Sacks Out, "FDA for AI" In: The Trump Administration's Quiet Reversal. David Sacks has formally concluded his tenure as White House AI and crypto czar, reaching the special government employee time limit. His departure coincides with the administration's pivot towards implementing an AI oversight framework he largely opposed. The Verge detailed Sacks's "crashed and burned" trajectory, noting that the administration, which began 2025 by revoking Biden's AI safety executive order, is now exploring an EO to establish a government vetting system for frontier AI models. Kevin Hassett, National Economic Council director, indicated to Bloomberg that the White House is considering reviewing AI models "like an FDA drug"—a remarkable shift in regulatory posture. This reversal, largely driven by Anthropic's Mythos and its security vulnerability disclosures, underscores the growing recognition that unregulated models pose national security risks. The absence of an immediate successor for Sacks leaves a vacuum in critical AI policy shaping. (The Verge, Bloomberg)

OpenAI and Anthropic Race to Own the Enterprise Stack. Both leading frontier labs are demonstrating aggressive strategies to dominate the entire AI deployment value chain, moving beyond mere API provision. PYMNTS and Reuters report OpenAI's $10 billion joint venture is pursuing the acquisition of three AI services firms, aiming to integrate consulting and implementation capabilities in-house. This move positions OpenAI to directly compete with its own "Frontier Alliances" partners such as Accenture and McKinsey. Simultaneously, Jefferies highlighted that both OpenAI and Anthropic are "going after AI services as an attempt to expand distribution ASAP." The parallel strategies of the Anthropic-Blackstone-Goldman JV and the OpenAI-TPG JV signal a direct assault on the enterprise IT services market. For established consulting and Indian IT firms, this implies an urgent need to adapt as model providers seek to capture a larger share of the enterprise wallet. (PYMNTS, Seeking Alpha)

Google and Meta Scramble to Build Personal AI Agents. Google and Meta are intensifying their internal efforts on consumer AI agents designed for autonomous task execution, as reported by Business Insider, FT, and The Information. Google's "Remy," a 24/7 personal agent integrated into Gemini across its ecosystem, is designed to act on user behalf, consolidating resources from its defunct Mariner browser agent project. Meta's "Hatch" is an agentic assistant coupled with a new AI-powered shopping feature for Instagram. The Decoder accurately characterizes this as a race between Google and Meta to catch up with Anthropic and OpenAI in the personal AI agent domain. The underlying strategic bet is clear: control of the user's daily task flow translates into ownership of the next computing platform. (Business Insider, Reuters, The Decoder)

X / Social Pulse

Murati's testimony commanded Thursday's discourse, with phrases like "chaos and distrust" trending alongside candid court appearance clips juxtaposed with her Met Gala imagery. This stark contrast—high fashion on Monday, hostile work environment testimony midweek—underscored the surreal dynamics of the ongoing trial. AI researchers widely noted that if Murati's claims about Altman's dishonesty on safety checks hold, it directly supports Musk's central argument regarding OpenAI's alleged deviation from its safety-first charter for commercial acceleration. Conversely, the revelation of Musk's prior attempt to integrate OpenAI into Tesla introduced a significant credibility challenge for his legal strategy, one that observers believe will be difficult to reconcile.

The news of Sacks's departure elicited a bifurcated response. While technologists who advocated for stronger regulation expressed satisfaction, the lack of an announced successor has raised questions about the continuity and direction of AI policy within the White House. Hassett's "FDA for AI" analogy is being scrutinized for its implications; an FDA-style approval process suggests pre-market regulatory hurdles, a fundamentally different paradigm from post-deployment accountability.

The EU's omnibus deal was met with mixed reception. Compliance teams within AI enterprises welcomed the extended timeline, but civil society organizations expressed concern that the 16-month delay could incentivize the deployment of harmful systems before rules are enforceable. The nudification ban, while broadly praised, was described by some as a compensatory measure within a broader deregulatory package.

One to Watch

CoreWeave's First Public Earnings – The AI Infrastructure Bellwether.

CoreWeave is scheduled to report its Q1 2026 results after market close today, with implications extending beyond its individual financial performance to the broader AI infrastructure market. Analysts anticipate $1.97 billion in revenue, a doubling year-over-year, alongside an expected loss of $0.89 per share, reflecting the substantial capital expenditures required for GPU-dense data center development. The stock has seen a 103% rally from its March lows, and options markets are pricing in an 18% post-earnings move, the widest implied range among AI infrastructure firms this quarter.

Key metrics to monitor include full-year 2026 revenue guidance (consensus: $12-13 billion), capital expenditure outlook ($30-35 billion targeted), and the composition of its order backlog. CoreWeave has indicated that nearly all of its 2026 capacity is pre-allocated, making demand signals for 2027-2028 critical for validating its aggressive expansion strategy. Thursday saw Core Scientific, a significant CoreWeave partner, disclose $208 million in Bitcoin sales during Q1 to finance its AI pivot, specifically its $10.2 billion, 12-year contract expansion with CoreWeave. A strong beat-and-raise from CoreWeave would affirm the structural, rather than cyclical, nature of AI compute demand, bolstering valuations across the AI infrastructure sector, from Nvidia to data center REITs. Conversely, a miss could trigger renewed scrutiny across the entire trade. (Benzinga, CoinDesk)

Quick Hits

  • Anthropic Commits $200 Billion to Google Cloud: A five-year agreement represents over 40% of Google's reported cloud revenue backlog, marking the largest single AI compute procurement deal disclosed to date. (Reuters)
  • Richard Dawkins Considers AI Consciousness: After a 72-hour dialogue with Anthropic's Claude, dubbed "Claudia," Dawkins published an essay exploring the possibility of AI consciousness, igniting debate among cognitive scientists and AI researchers. (The Guardian)
  • IBM Positioned as "Full-Stack Enterprise AI Operator": Following Think 2026, Morgan Stanley upgraded IBM's standing, citing its Sovereign Core, Enterprise Advantage, and Concert operations platform as drivers for a strengthened enterprise AI position. (Seeking Alpha)
  • AI Captures 95.2% of Insurtech Funding: Q1 2026 saw AI-centric companies secure $1.55 billion across 68 deals in insurtech, representing the highest concentration of AI investment in any vertical sector, according to Gallagher Re. (The Insurer)
  • CNN Analyzes Pentagon AI Legal Framework: A deep dive examines the legal limitations governing the U.S. military's deployment of AI systems in active conflict zones, specifically referencing the ongoing Iran campaign. (CNN)

The convergence of internal corporate governance, regulatory shifts, and infrastructure investment is accelerating. Murati’s testimony provides critical insider validation for Musk’s claims, sourced from a former OpenAI CTO. Concurrently, the EU’s 16-month delay on its own AI Act, paired with Sacks’s departure and Hassett’s "FDA for AI" proposal, signals a period of profound regulatory flux across major economic blocs. Amidst this, capital continues to flow: Anthropic's $200 billion commitment to Google Cloud, OpenAI's strategic acquisitions, and CoreWeave's impending earnings report underscore relentless investment. The central question remains whether robust governance mechanisms can be effectively established before the next generation of frontier models fundamentally reshapes market dynamics and global incentives.

Sources

Lock in. M. mazen@thorterminal.com

Read more