The Omniscient Bulletin · 2026-05-03
The Omniscient Bulletin — May 3, 2026
Pentagon AI deals, the Anthropic paradox, the Musk v. Altman bombshell, SoftBank's Roze spinout, Google's $40B bet, and Grok 4.3's aggressive pricing.
The Pentagon is quietly building a multi-vendor AI stack cleared for its most sensitive classified networks — and the vendor list now runs to eight. That expansion is the lead today, but the bigger story underneath it is structural: the same company the Defense Department tried to freeze out (Anthropic) just attracted $40 billion from Google, and its restricted Mythos model is already in use at the NSA. Meanwhile, a federal trial threatens to upend OpenAI's march toward a trillion-dollar IPO — with the bombshell revelation that Musk's own xAI trained on OpenAI's models. SoftBank is betting $100 billion that the next act is AI-built data centers, and Grok 4.3 is cutting prices sharply to grab developer share.
Policy
Pentagon clears 8 firms to run AI on classified networks, freezing out Anthropic
The Defense Department signed agreements with Nvidia, Microsoft, AWS, Google, OpenAI, SpaceX, Oracle, and Reflection AI to deploy models on Impact Level 6 and 7 classified networks. Oracle was added hours after the initial announcement. Pentagon CTO Emil Michael took a veiled shot at Anthropic, saying the DoD had learned it was "irresponsible to be reliant on any one partner" that "didn't really want to work with us." Anthropic won a federal injunction in March after the DoD tried to designate it a supply-chain risk over guardrail disagreements.
Industry
Musk admits xAI distilled OpenAI's models as trial threatens OpenAI's IPO
Week one of Musk v. Altman produced a courtroom bombshell: under cross-examination, Elon Musk admitted that xAI "partly" distilled OpenAI's models to build Grok — the same practice OpenAI condemned when DeepSeek did it in 2025. Musk is seeking to remove Sam Altman and Greg Brockman and unwind OpenAI's for-profit restructuring. The outcome could derail OpenAI's race toward an IPO at a valuation approaching $1 trillion. Week two begins today with Stuart Russell testifying, followed by Brockman.
Frontier
xAI launches Grok 4.3 with aggressive pricing and a new voice cloning suite
xAI released Grok 4.3 on April 30 with a 1M-token context window and API pricing 40–60% below Grok 4.2 — a direct bid for developer share from OpenAI and Anthropic. The model features always-on reasoning and scores 53 on the Artificial Analysis Intelligence Index, a jump over 4.2 but still trailing the top offerings. xAI also shipped a voice cloning suite that replicates voices from short audio samples. No system card, model card, or press release accompanied the launch — continuing xAI's pattern of skipping the transparency documentation its rivals now treat as standard.
Industry
SoftBank prepares $100 billion Roze spinout to build AI data centers with robots
SoftBank is planning to list a new entity called Roze on U.S. markets at a $100 billion valuation, with a potential IPO in the second half of 2026. Roze's focus is specific: deploying autonomous robot fleets to build AI data centers faster and more cheaply. The venture could also bundle energy, land, and infrastructure assets alongside ABB Robotics, a global automation supplier SoftBank agreed to acquire last year. CEO Masayoshi Son is driving the effort partly to offset the firm's $30 billion-plus committed to OpenAI. Some executives have called the timeline ambitious.
Industry
Google commits $40 billion to Anthropic as the Pentagon tries to shut the company out
Google is committing up to $40 billion to Anthropic — $10 billion immediately at a $350 billion valuation, $30 billion contingent on performance targets — making it the largest single investment in Anthropic's history. Google Cloud is also providing 5 gigawatts of TPU capacity over five years. Amazon added a fresh $5 billion the same week. The timing is striking: while the Pentagon publicly froze Anthropic out of classified networks, the company secured what amount to the two largest investments in frontier AI history.
Frontier
Anthropic's restricted Mythos model is already at the NSA — despite the Pentagon ban
Anthropic's Mythos Preview, restricted to roughly 40 organizations over concerns about its offensive cybersecurity capabilities, is being used by the NSA to scan for exploitable vulnerabilities. The UK's AI Security Institute has also confirmed access. The disclosure deepens the contradictions in the government's posture: the DoD is fighting Anthropic in court while the NSA quietly uses its most powerful — and most dangerous — model. Anthropic CEO Dario Amodei met with White House chief of staff Susie Wiles last week; the administration called the meeting productive.