The Omniscient Bulletin · 2026-05-05
The Omniscient Bulletin — May 5, 2026
Frontier AI gets a government minder, Five Eyes pump the brakes on agents, and $950M lands on enterprise AI
Policy
Microsoft, Google DeepMind, and xAI sign Commerce Dept deals for pre-release model vetting
The Commerce Department's Center for AI Standards and Innovation announced Tuesday that Microsoft, Google DeepMind, and xAI have agreed to let the government run security evaluations on their frontier models before public release. Existing agreements with OpenAI and Anthropic were simultaneously renegotiated under Trump administration priorities. The moves follow NYT reporting that the White House is drafting an executive order to formalize pre-release model review — a significant policy reversal from the deregulatory posture the administration held just months ago.
Agents
Five Eyes intelligence agencies warn: slow down on agentic AI or risk serious security breaches
CISA, NSA, and their counterparts from the UK, Canada, Australia, and New Zealand jointly published guidance Friday urging organizations to treat agentic AI deployments with extreme caution. The document catalogues 23 distinct risks and more than 100 recommended controls, warning that each new tool or data source an agent touches widens the attack surface. Until security standards mature, the agencies say, enterprises should assume agents will behave unexpectedly — and design systems that can be reversed, not just monitored.
Industry
Sierra raises $950M at $15B+ valuation as enterprise AI agent spending accelerates
Bret Taylor's AI customer-service startup Sierra closed a $950 million Series E led by Tiger Global and GV, pushing its valuation above $15 billion in under six months. The company claims more than 40% of Fortune 50 companies now run on its platform, with agents handling billions of interactions from insurance claims to mortgage refinancing. Sierra also recently launched Ghostwriter, a tool that autonomously builds and deploys specialized agents from natural-language descriptions.
Policy
NHS England orders GitHub repositories made private amid AI hacking fears, reversing open-source commitments
Anthropic's Mythos is reshaping software security policy well beyond tech circles. NHS England has instructed technology leaders to make nearly all its GitHub repositories private, citing fears that frontier AI models could use public code to find exploitable vulnerabilities at scale. The move contradicts long-standing NHS commitments to open-source software for publicly funded code — and has drawn sharp criticism from open-source advocates who argue it trades collaboration for security theater.
Compute
Peter Thiel backs $140M wave-powered ocean AI data centers in bid to break land and grid constraints
Portland-based startup Panthalassa closed a $140 million Series B led by Peter Thiel. Its floating nodes generate power from wave turbines, use seawater for cooling, and transmit data via low-Earth-orbit satellites — entirely off-grid. Investors including John Doerr, Marc Benioff's TIME Ventures, and Max Levchin's SciFi Ventures also participated. The bet reflects mounting pressure on conventional data centers: land scarcity, power grid bottlenecks, and water-use constraints are pushing AI infrastructure into increasingly unconventional territory.
Frontier
OpenAI quietly shelved plans to spin out its robotics and hardware units ahead of IPO
Sam Altman explored spinning off OpenAI's robotics and consumer-hardware divisions late last year under an Alphabet-style structure to give them independent funding. The plan collapsed because both units would likely remain on the consolidated balance sheet regardless, offering no financial separation. The WSJ reports OpenAI has shelved the talks entirely, refocusing on a coding-and-enterprise superapp as its core IPO narrative — trimming side ventures to sharpen the story it tells public-market investors.