AI Formally Verifies Signal's Cryptography at Machine-Code Level

Anthropic IPO oversubscribed multiple times, targeting $90.4B raise with Amazon and Google holding 25%+ combined

Anthropic's initial public offering has drawn multiple times oversubscribed demand, with the company targeting a $90.4B raise. Amazon and Google collectively hold stakes worth potentially $250B+ and $140B respectively. Dario Amodei claimed AI will make cancer, Alzheimer's, and heart disease "relics of the past" in under a decade, signaling massive institutional conviction in Anthropic's competitive position against OpenAI and validating the AI model market's scale.

Constellation Software running 1,600+ AI experiments across portfolio with distributed learning model as competitive moat

Management disclosed that nearly all AI learning happens within individual portfolio companies rather than at head office, with results traveling rapidly across the group via established knowledge-sharing mechanisms. This distributed experimentation model is difficult to replicate and could translate to organic growth and margin improvements if best practices prove durable and broadly applicable.

Nvidia board moves to oust Jensen Huang amid affair with alleged Chinese intelligence operative

Huang confirmed a relationship with Megaspeed executive Alice Huang, who allegedly cooperated with China's Ministry of State Security, triggering board action and severe reputational damage to Nvidia in Washington. This directly undermines Nvidia's position ahead of critical US-China chip negotiations at APEC in Shenzhen next week and signals potential intensification of export controls.

AlphaProof Code: Verifying Signal's Security at the Machine-Code Level

Google DeepMind published a technical report on AI-assisted formal verification of cryptographic protocol implementations. Main takeaways:

• AlphaProof Code framework uses multi-agent AI to generate 6.5M lines of machine-checked Lean proofs verifying Signal protocol security properties at compiled AArch64 machine-code level, first such proof for unmodified production cryptographic implementation.
• Specification writing (18K lines trusted code) remains human-guided bottleneck; proof generation (6.5M lines) now automated by AI agents whose output is kernel-verified, shifting labor from proof engineering to specification design.
• Verification covers PQXDH handshake and Triple Ratchet establishing forward secrecy, post-compromise security, and quantum-resistant handshake secrecy; mutation testing re-introduced two historical libsignal bugs and proved violations, validating specification strength.
• Demonstrates formal verification of real-world software now tractable via AI; prior comparable efforts (seL4 microkernel, CompCert compiler) required 6-20 person-years; this work suggests dramatic cost reduction for critical infrastructure security assurance.

Trump Administration Standoff Over Nvidia Chip Export Controls to China; Lutnick Blocks Megaspeed Sanctions Despite Espionage Scandal

Axios reported on internal Trump administration friction over export controls on Nvidia chips diverted to China, specifically targeting Megaspeed, a Singapore-based AI firm at the center of a suspected espionage scandal. Main takeaways:

• Commerce Secretary Howard Lutnick has repeatedly blocked three separate Trump-approved decisions (May, August, Nov 5) to impose export controls on Megaspeed despite pressure from Treasury, Defense, and National Security officials
• A classified memo documented Jensen Huang's relationship with Alice Huang, a Megaspeed employee suspected by U.S. intelligence of being a Chinese agent, shared with Trump and cabinet officials
• Trump authorized controls after consulting Elon Musk, but Lutnick proposed delays post-decision; implementation has not occurred despite explicit presidential directive
• Lutnick and Nvidia have lobbied for looser export controls, arguing that allowing sales of older chips (H200) to China prevents Chinese self-sufficiency and maintains U.S. leverage