The Omniscient Bulletin · 2026-05-13
The Omniscient Bulletin — May 13, 2026
Altman's financial conflicts take center stage in cross-examination, AISI finds a universal jailbreak in GPT-5.5-Cyber, and Jensen Huang boards Air Force One at the last minute for Beijing.
Wednesday's signal turns on three stories that sharpened overnight. Musk's attorney spent Tuesday afternoon pressing Altman on a sprawling web of personal investments - a $1.65B Helion stake, a $632M Stripe position, and more than $2B in total across companies doing business with OpenAI - putting the conflict-of-interest question at the center of the trial as closing arguments approach. UK AISI's evaluation of GPT-5.5, published April 30, lands with new weight given the Daybreak launch: the institute found a universal jailbreak in six hours of expert red-teaming, complicating OpenAI's cyber-defense positioning. And Nvidia's Jensen Huang boarded Air Force One at a refueling stop in Alaska after a last-minute Trump call - the trip that was supposed to happen without him suddenly becomes the most consequential diplomatic moment for chip export policy in months.
Industry
Molo presses Altman on $2B investment web as cross-examination concludes; closing arguments expected this week
Musk's attorney Steven Molo used Tuesday's cross-examination to build a conflict-of-interest case around Altman's personal portfolio: a one-third Helion Energy stake (~$1.65B), $632M in Stripe, and more than $2B total across companies doing business with OpenAI. Molo also pressed Altman on the $200M Reddit data deal, negotiated while he held a significant Reddit position. Altman said he recused himself from relevant board decisions; asked directly about his character, he replied: "I believe I am an honest and trustworthy business person." Closing arguments are expected as soon as this week.
Frontier
AISI found a universal GPT-5.5 jailbreak in six hours - a complicating footnote to Daybreak's launch
UK AISI's April 30 evaluation takes on new relevance after Monday's Daybreak launch. GPT-5.5 scored 71.4% on Expert-level cyber tasks, edging Mythos Preview's 68.6%, and completed AISI's 32-step corporate-network simulation in 2 of 10 attempts vs. Mythos's 3 of 10. The safeguards finding cuts the other way: expert red-teamers built a universal jailbreak covering every malicious cyber query in the test set in six hours. OpenAI updated its safeguard stack in response, though AISI could not verify the final configuration.
Compute
Trump calls Huang at the last minute; Nvidia CEO boards Air Force One for Beijing as H200 hopes ride along
Huang was not on the original delegate list for Trump's China summit - his absence drew pointed media coverage about chip-sales friction. Trump called him directly after seeing the reports; Huang boarded Air Force One at an Alaska refueling stop. The summit runs Thursday and Friday in Beijing alongside Musk, Tim Cook, and Qualcomm's Cristiano Amon. Huang has lobbied for months to ease H200 export restrictions. Experts are skeptical a deal is imminent, but his presence turns the trip from a policy statement into an active commercial negotiation.
Labor
Meta's first 2026 layoff wave hits in seven days as restructuring scope sharpens
Meta's May 20 cut is now a week out: roughly 8,000 positions eliminated, 6,000 open requisitions cancelled, surviving teams reorganized into AI-focused pods. Structural, not performance-based, and funded by a $115-135B AI infrastructure budget that is 73% larger than 2025's. A second wave is planned for H2. Worth tracking which orgs absorb cuts and which expand; the gap is the company's revealed bet on which AI work is durable. Executive comp tied to a $9T-by-2031 market-cap target keeps the structural framing intact.