OpenAI Shares Stuck, Billions Chase Anthropic.
TL;DR
- Political Engagement: Anthropic established its first employee-funded PAC, AnthroPAC, signaling a strategic shift towards direct political influence as AI governance becomes a key midterm issue.
- Data Breach Fallout: A supply-chain compromise affecting LiteLLM led to Mercor exposing four terabytes of AI training data from major players, prompting a class-action lawsuit and Meta freezing its partnership.
- Content Innovation: Netflix open-sourced VOID, an advanced physics-aware AI model capable of intelligently removing objects from video and regenerating the surrounding scene dynamics.
- India's AI Ascendance: Sarvam AI is poised to close the largest AI funding round in India's history, securing $300-350 million to develop localized, voice-first AI solutions.
- Market Anticipation: SpaceX confidentially filed for an IPO at a $1.75 trillion valuation, positioning its combined entity with xAI to potentially absorb significant public market AI appetite ahead of rivals.
Lead Story: Anthropic's Strategic Pivot to Political Influence
Anthropic filed paperwork Friday to launch AnthroPAC, marking the company's first direct foray into federal political action. This employee-funded PAC will channel voluntary contributions, capped at $5,000 annually per person, to candidates involved in AI policy, operating under a bipartisan board and full FEC disclosure.
The timing of this move is strategically significant, coinciding with Anthropic's ongoing lawsuit against the Pentagon, its preparations for an IPO, and the escalating prominence of AI governance in the midterm election cycle. Collective spending by AI companies on the 2026 races has already surpassed $300 million, underscoring the high stakes.
The "bipartisan" framing immediately faced scrutiny. Breitbart flagged that the vast majority of Anthropic employees' historical political donations favored Democrats, while the Washington Examiner questioned the PAC's ability to genuinely cross the aisle amidst the company's legal challenge against the administration.
Axios reports this initiative signifies a fundamental shift: Anthropic is evolving from a research entity navigating policy to an active participant shaping the legislative landscape for AI. With the Ninth Circuit appeal due April 30 and an October IPO target, each political investment is directly tied to core business objectives.
The Hill notes that while other major AI players like OpenAI, Google, and Microsoft have established PACs, Anthropic's launch occurs under uniquely adversarial conditions. No other AI company has initiated a PAC while simultaneously engaged in litigation against the federal government, highlighting a distinct strategic posture.
In Other News
Meta freezes Mercor after supply-chain attack exposes four terabytes of AI training data. The $10 billion AI data startup confirmed to Business Insider that a breach, linked to a LiteLLM supply-chain compromise, exposed sensitive training methodologies used by Meta, OpenAI, and Anthropic. The hacking group Lapsus$ claimed responsibility, leading to a class action filed April 1 alleging over 40,000 individuals, including contract employees, now face identity theft and fraud from exposed SSNs and PII.
Netflix open-sourced VOID, a physics-aware AI that erases objects from video. VOID goes beyond simple object removal, intelligently regenerating how the remaining scene would physically behave without the excised element. Built on CogVideoX, The Register notes this marks Netflix's first open-source AI model, now available on Hugging Face with open weights. In comparative user studies, VOID achieved a 64.8% preference rate over Runway (18.4%), underscoring its advanced capabilities.
Sarvam AI is closing India's largest-ever AI funding round. The Bengaluru-based startup is nearing completion of a $300-350 million funding round at a $1.5 billion valuation, with Bessemer leading and participation from Nvidia, Amazon, and Prosperity7. Sarvam's focus on AI models optimized for India's 22 official languages, particularly voice-first interaction, represents a strategic bet on serving the next billion AI users who may not primarily engage in English text.
X / Social Pulse
AnthroPAC's debut cleaved opinion along familiar political lines. Tech-aligned accounts framed it as an inevitable defense against perceived governmental overreach, while conservative commentators interpreted it as confirmation of Anthropic's inherent political agenda. Netflix's VOID generated genuine enthusiasm within VFX and post-production communities, lauded as a potential disruptor capable of consolidating complex pipelines into a single model. The confidential SpaceX IPO filing fueled broader discussions about "2026: the year public markets truly meet AI," with analysts cautioning that SpaceX's $1.75 trillion valuation could significantly absorb institutional demand before OpenAI or Anthropic even reach the S-1 stage.
One to Watch
The mega-IPO queue is taking shape — and SpaceX just cut in line. SpaceX confidentially filed for a June IPO at a $1.75 trillion valuation, targeting up to $75 billion in proceeds. This offering, covering the combined SpaceX-xAI entity following February's all-stock acquisition, could set a new record. Should SpaceX precede its AI-centric rivals to market, its immense scale could significantly capture institutional and retail AI investment, potentially impacting the valuations and market reception of OpenAI (~$1T target) and Anthropic ($400-500B target) upon their eventual listings.
Quick Hits
- Anthropic's Model Context Protocol (MCP) crossed 97 million installs in March, representing the fastest adoption curve for any AI infrastructure standard to date, with broad industry integration.
- OpenAI fully retired GPT-4o on April 3 across all plans, completing a phased sunset initiated in February, as daily usage had declined to 0.1% with GPT-5.2 and 5.4 dominating.
- Anthropic published a multi-agent harness architecture designed for long-running autonomous development, a GAN-style loop where evaluator agents provide iterative critique across design, originality, and functionality in 4-hour sessions.
- AI stocks that performed in 2025 are not replicating that success in 2026 — Motley Fool reports a market shift towards critical evaluation of actual AI revenue versus speculative hype, particularly pressuring SaaS valuations.
- Gemma 4 benchmarking indicates Google's open-weight family is now competitive with closed-source models within the 2B-31B parameter range, setting the stage for a critical test with the anticipated DeepSeek V4 release.
Anthropic establishing a PAC while simultaneously litigating against the government and preparing for an IPO distills the current strategic landscape. AI companies are no longer simply developing technology; they are actively constructing the political and market architecture necessary to secure its future.
Sources
AnthroPAC: TechCrunch, Axios, The Hill, Decrypt, Washington Examiner, Breitbart, Implicator
Meta-Mercor Breach: Business Insider, Fortune, TechCrunch, ClaimDepot, Security Magazine, The Next Web
Netflix VOID: MarkTechPost, The Register, The Decoder, Hugging Face
Sarvam AI: Bloomberg, Outlook Business
MCP & Models: AI Unfiltered/MCP, OpenAI/GPT-4o Retirement, InfoQ, Epsilla, Motley Fool, Colleges Simplified/Gemma 4
Lock in. M. mazen@thorterminal.com