|
|
|
A leak firing and a broken-culture resignation hit the same team in one week.
|
|
On October 1, OpenAI fired safety researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni for sharing confidential infrastructure details with an outside AI-safety group, per The Hacker News, citing the Wall Street Journal and Bloomberg. On October 3, David Robinson, who says he led safety reports for OpenAI's major launches over three and a half years, quit and called the company's culture "broken" in an essay for The Atlantic, per TechCrunch. Keep the two apart. The firings are a confidentiality breach; Robinson's essay argues OpenAI lacks the redundancy of safety-critical industries like nuclear plants and airports. It follows ex-OpenAI and Anthropic researcher Jacob Coxon's September 9 exit, when he warned the labs were "gambling with our lives," per TechCrunch. Nothing about the API changed this week. If OpenAI sits on your critical path, log both threads in your vendor-risk file and confirm your fallback model actually runs. Read the full timeline.
|
|
|
|
•
|
Google paused its open-source bug bounty over AI slop. Google froze its open-source Vulnerability Rewards Program as of October 1, citing "a significant rise in automated submissions, the vast majority of which are not valid," per TechCrunch. It plans to reopen in Q1 2027. Run a bounty yourself? Require a working reproduction before any human triages a report, because the flood that closed Google's program will not stop at Google.
|
|
•
|
GitHub retired four Copilot models on October 2. Gemini 3.5 Flash and 3.6 Flash now map to Gemini 3.8 Flash, Kimi K2.7 Code to Kimi K3, and Claude Opus 4.7 to Claude Opus 5.5, per GitHub's changelog. The model picker swaps them automatically, but GitHub says pinned workflows and integrations need updating. Grep your CI configs for the old IDs today, not after a job fails.
|
|
•
|
Barclays published real adoption numbers for Claude. Barclays wants 50 percent of its developers on Claude by the end of 2026, its Colleague Knowledge Assistant has served over 16,000 UK staff and more than 1 million searches since 2025, and Claude routes roughly 120,000 emails a day in Global Markets, per Anthropic's newsroom. Building the case for a rollout inside a regulated company? These are your comparables, though a vendor published them, so treat them as a ceiling.
|
|
•
|
Anthropic is betting the enterprise bottleneck is people. Claude Frontier Academy, a $100 million program launched October 2, aims to train 10,000 "Frontier Deployed Engineers" by the end of 2027, per Anthropic's newsroom. Anthropic is telling you where enterprise deployments stall: integration and operations talent. If you can wire a model into a legacy system, that skill now has a well-funded sponsor and, soon, a credential.
|
|
•
|
Trump launched a "Super Intelligence Force" with a 120-day clock. Announced October 4 on Truth Social and chaired by National Intelligence Director Jay Clayton, with FTC Chair Andrew Ferguson among the vice chairs, the task force owes a report on AI risks and opportunities in 120 days, per TechCrunch. Its stated goal is to avoid regulation that "would stifle innovation." The report lands around early February 2027; plan compliance work against the rules that already bind you, not this one.
|
|
|
|
|
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
|
|
|
Affiliate disclosure
·
Pondero earns commissions on some links. This does not affect our editorial picks.
|
|