Amodei, Bessent Meet On Mythos Truce.

Mazen Al Sabbagh

— 7 min read
Share

TL;DR

  • Diplomatic Breakthrough: Anthropic CEO Dario Amodei engaged directly with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent, marking the first West Wing interaction since the Pentagon blacklisting. Both sides termed the session "productive," with Bessent's presence underscoring Mythos access for financial systems as a critical priority.
  • OpenAI Leadership Restructure: OpenAI saw the simultaneous exit of product chief Kevin Weil, Sora head Bill Peebles, and enterprise CTO Srinivas Narayanan. This signals a definitive end to the company's "side quest" initiatives, shifting focus back to core model development amidst mounting investor scrutiny over Sam Altman's external ventures.
  • xAI's Strategic Growth: Grok 4.3 Beta, a 0.5T-parameter model with advanced multimodal capabilities, debuted for SuperGrok Heavy subscribers, with a 1T version imminent. Critically, xAI is now supplying GPU infrastructure to Cursor for its model training, indicating a significant play in AI compute architecture.
  • Cursor's Exponential Valuation: AI coding firm Cursor is in advanced discussions to secure $2 billion at a $50 billion valuation, nearly doubling its market perception in weeks. This investment, with backing from a16z, Thrive, and Nvidia, cements Cursor's position as a foundational interface for AI development.
  • Embodied AI's Public Test: Over 300 humanoid robots will compete in a Beijing half-marathon tomorrow, with 40% operating autonomously. This event represents the most extensive public assessment of embodied AI's locomotion and commercial readiness to date.

Lead Story: The Thaw Begins — Amodei Walks Into the West Wing, Walks Out With "Productive" on the Record

The engagement transpired. Reuters confirmed Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles Friday evening to discuss Mythos access. This marks the administration's first direct West Wing dialogue since the Pentagon blacklisted Anthropic in February. The critical presence of Treasury Secretary Scott Bessent in the room, as reported by Axios, alongside the blunt assertion that "Every department other than War wants to use Anthropic," frames this meeting within a wider strategic imperative.

Both parties characterized the session as "productive" and "constructive," with future collaborative avenues under consideration, according to The New York Times. This represents a notable policy pivot. Politico's analysis highlighted the journey from President Trump's earlier condemnation of Anthropic as "woke" to the current structured negotiations. The BBC accurately summarized the situation: the firm's technology is becoming indispensable, even for the U.S. government.

The core motivation remains financial. Secretary Bessent's direct involvement underscores that securing Mythos access for the banking sector and the Treasury's cybersecurity operations is now a co-equal national priority, aligning with the earlier Bloomberg report that the White House was already facilitating internal agency access. This meeting formalizes the strategic momentum.

Further validating Anthropic's stance, Business Insider noted Steve Bannon, at the Semafor World Economy Summit, supported Anthropic's refusal of the Pentagon's initial terms, advocating for a deeper understanding of AI's military applications. Concurrently, AI pioneer Yoshua Bengio told Fortune that Mythos' capabilities illuminate "AI's concentration of power," necessitating "urgent international cooperation." This positions Anthropic's controlled deployment not merely as a company decision, but as a symptom of a broader architectural challenge.

While no formal agreement materialized, and the Ninth Circuit oral arguments on May 19 persist as a legal contingency, the political resolution track is demonstrably outpacing judicial processes.

In Other News

OpenAI’s Strategic Realignment Underway. Three senior executives departed OpenAI concurrently. TechCrunch reported the exit of product chief Kevin Weil and Sora head Bill Peebles, with The Next Web confirming the third as enterprise CTO Srinivas Narayanan. Bloomberg corroborated these exits. Sora is slated for closure, and the science team is being integrated into core operations. CryptoBriefing noted a 12-point jump in Polymarket odds for GPT-5.5 by June 30, implying market expectation for an accelerated model release. Concurrently, Gizmodo highlighted institutional investor concerns regarding Sam Altman’s personal portfolio, specifically his World (Worldcoin) identity venture — which just secured partnerships with Zoom and Tinder — as the firm approaches its IPO.

xAI's Compute Dominance Emerges. xAI formally released Grok 4.3 Beta on April 17 to SuperGrok Heavy subscribers, priced at $300/month. This 0.5-trillion-parameter model features advanced long-context processing, video understanding, and robust PDF/slide generation. Elon Musk confirmed on X that the 1T version will complete training within five days. New speech-to-text and text-to-speech APIs also debuted with competitive pricing. Testing Catalog previewed Grok Build and Grok CLI, slated for next week, signifying xAI's expansion into comprehensive developer tooling. The most strategic development, however, came with TechWire Asia's report that xAI is now providing GPU infrastructure to Cursor for training its Composer 2.5 coding models. This arrangement—where a $50 billion valuation company trains on Musk’s compute, while Nvidia invests in both Cursor's round and the Terafab—underscores a reshaping of the core AI supply chain.

Cursor's Enterprise Value Surges. The AI coding startup, a notable disruptor in the IDE market, is poised to nearly double its valuation. Bloomberg and TechCrunch reported Cursor is negotiating a $2 billion raise at a $50 billion valuation, an ascent from $29.3 billion within weeks. a16z and Thrive are returning as anchor investors, with Nvidia also participating. Current annualized revenue has reached $2 billion. For a company that introduced Cursor 3 "Glass" with an agent-first architecture on April 2, this valuation transcends a mere code editor, positioning it as the emerging primary interface for developers interacting with diverse AI models.

Infrastructure Bottlenecks Impede AI Expansion. The foundational physical infrastructure for the AI boom is showing significant strain. Ars Technica and the Financial Times cited SynMax satellite and drone analysis, revealing that 40% of planned U.S. AI data center construction projects are experiencing delays. Key factors include labor shortages, material constraints, and power delivery bottlenecks. Microsoft and OpenAI projects are among those flagged. While Tom's Hardware observed corporate denials of schedule issues, satellite intelligence indicates otherwise. Concurrently, residents near Musk's xAI Terafab expansion in Memphis are protesting, terming it a "death sentence" due to noise, power grid strain, and environmental impact — giving a local dimension to the broader infrastructure backlash.

X / Social Pulse

Opus 4.7 Backlash Intensifies. The market reaction to Claude Opus 4.7 is crystallizing. Business Insider reported on "The Claude-lash," noting token consumption issues and user frustration. every.to's assessment articulated it sharply: "More precise, more literal, best coding model we've tested on well-specified tasks — but it won't fill in the gaps for you anymore." The emergent consensus, per Apiyi.com, transitioned rapidly from viewing 4.7 as a "comprehensive upgrade" to a "selective upgrade" within two days of release.

Bengio Joins Mythos Governance Discourse. Turing Award laureate Yoshua Bengio told Fortune that Mythos exemplifies "the problem of AI's concentration of power," demanding urgent international collaboration. This, coupled with Steve Bannon's public endorsement of Anthropic's position at Semafor, positions the Mythos governance debate squarely across the entire political and ideological spectrum, from the populist right to the academic safety establishment.

Congressional Anxiety Over AI Deepens. AP News reported that lawmakers at a bipartisan AI roundtable used the phrase "fears of destruction" during discussions with AI executives and academics. Fast Company observed the mood to be palpably more anxious than any prior AI congressional hearing.

One to Watch

Embodied AI: Beijing's Half-Marathon as a Real-World Prove Out. Reuters confirmed that 112 teams, including Alibaba, Honor, Unitree, and AGIBOT, will deploy over 300 humanoid robots in a 21km race through Beijing's E-Town on April 19. Notably, approximately 40% will operate with complete autonomy. Seeking Alpha reported investors are closely monitoring for "breakout" performances, viewing this as a critical proxy for embodied AI's commercial readiness. Reflecting global momentum, Siemens recently trialed an Nvidia-powered humanoid robot for autonomous logistics at its Erlangen, Germany electronics factory — a significant European industrial deployment. AGIBOT, today at its Partner Conference, separately declared 2026 as "Deployment Year One". Upcoming, Google I/O is scheduled for May 19, coincident with the Ninth Circuit Anthropic oral arguments.

Quick Hits

  • Meta's Workforce Restructuring: Reuters reported exclusively Meta targets May 20 for an initial wave of 8,000 layoffs, with up to 16,000 total across two rounds expected in 2026, per India Today.
  • Alibaba's Efficiency Edge: The Decoder confirmed Alibaba's Qwen3.6-35B model outperforming Google's Gemma 4-31B on agentic coding benchmarks, utilizing only 3B active parameters, signaling continued efficiency gains from Chinese open-source initiatives.
  • FISA 702 Short-Term Extension: After long-term renewal attempts failed, NPR reported FISA Section 702 was extended to April 30. Senators Lummis and Wyden previously warned of "unprecedented mass surveillance" at the intersection of 702 and AI capabilities.
  • Anthropic Disrupts Design Market: Anthropic launched Claude Design, a generative tool transforming text prompts into interactive prototypes. This announcement led to a 7% drop in Figma stock, as reported by TechCrunch and Sherwood News.
  • Worldcoin's Expanding Reach and Scrutiny: Axios reported Sam Altman's World (Worldcoin) secured partnerships with Zoom and Tinder for AI-vs-human identity verification. This coincides with Gizmodo's reporting on investor concerns regarding potential conflicts of interest from Altman's personal ventures ahead of OpenAI's IPO.

The White House meeting concluded without a signed accord. Yet, its sheer occurrence — Treasury's presence, the "productive" record, Bannon's unexpected validation, Bengio's framing of a governance crisis — signals a profound shift in power dynamics. Weeks prior, Anthropic's market viability was questioned amidst a Pentagon blacklisting. Today, the strategic query shifts: can the Pentagon effectively operate without Mythos access when critical sectors of government and industry increasingly demand it? As this political theater unfolds, Anthropic demonstrated market impact with Claude Design's launch and Figma's subsequent 7% stock decline; xAI strategically deployed Grok 4.3 while securing a foundational role as Cursor's compute provider; and OpenAI underwent a significant executive purge. The entity that once challenged governmental directive now witnesses the very institutions and competitors align, perhaps inadvertently, with its initial thesis. The market is adjusting its architecture.

Sources

Lock in. M. mazen@thorterminal.com

Read more