CI secrets stolen across 61k repos; Copilot's retention catch; Nvidia's board moves on Huang

🔥The day's one thing

🏢How organisations are changing

• Several major 2026 layoffs name AI as a driver. Business Insider's running 2026 tracker counts job cuts at dozens of major employers, and AI is named as a driver at several — Block, Coinbase and Standard Chartered all cited it. The numbers for board conversations: Atlassian is cutting 1,600 people, about 10% of staff, as it invests in AI; Cloudflare is cutting 20% of its workforce after internal AI use climbed more than 600% in three months; Barclays is cutting 5,000, with 60% from support functions including fewer developers. Atlassian took $225–236 million in restructuring charges for its cut, announced March 11. Business Insider

  “We fundamentally believe people and AI create the best outcomes. Our approach is not 'AI replaces people,'”

• The 100% AI mandate, one year on. A Hacker News thread on a post claiming a year-old mandate to write all code with AI left a codebase nobody understands is the sharpest collection of rollout-failure evidence this week — though the thread itself flags that the original post may be AI-generated, so read it as anecdote, not study. The most concrete account in it: a developer's friend at a large F100 company now spends 10-hour days feeding Claude and says he is learning nothing. The thread's usable warning for anyone weighing a mandate-style rollout: the cost shows up as collective understanding of the system, which is exactly what a mandate never measures. Commenters also note nobody fully understood large codebases before AI either — the credible middle position is measuring comprehension, not assuming it. reddit.com · from the Hacker News discussion

• AWS Emulator — a free local stand-in for more than 200 AWS services, released November 11 under CC0, that a coding agent can test its own work against — and Amazon says agents wrote and checked every line, working from over 400 API models. The lesson for engineering leaders is in the post itself: when agents check their own work, engineers move to writing the specifications and building the tests and merge gates that decide when that output counts as correct. Kiro turned one prompt into a passable first SQS emulator in about 15 minutes. All Things Distributed

🧾Vendors, pricing and risk

• Claude Fable 5.2 in GitHub Copilot — is now generally available — but unlike every other Claude model in Copilot, it needs Anthropic to retain prompts and outputs to run its safety classifiers, and the policy is off by default for Copilot Business and Enterprise until an administrator turns it on. Treat enabling it as a data-terms decision, not a settings checkbox: your engineers' prompts stay with Anthropic, and eligible enterprises keep zero-retention access to Fable models only through the end of the calendar year, after which Anthropic's Enterprise Frontier Safeguards program governs it. Enabling the policy formally acknowledges the retention requirement, per GitHub's changelog. GitHub Blog

• GitHub Agentic Workflows — is now generally available: coding agents run inside GitHub Actions on issue triage and CI-failure analysis, start read-only, and have their changes applied by separate jobs holding the write tokens, with a threat-detection scan before anything lands. Spend is capped by default — AI Credits cost $0.01 each, with 1,000 credits ($10) per run. One tester at freenet found even the default too generous for triage and capped runs at 100 credits, 15 turns and 10 minutes. GA announced November 11; the default Copilot engine requires a Copilot licence. InfoQ

📰News the board will ask about

• Nvidia's board is reportedly preparing to oust Huang. The Wall Street Journal reports Nvidia's board is preparing to remove CEO Jensen Huang, after the New York Times — citing five employees — reported a second affair, this one with vice president of healthcare Kimberly Powell, who reports directly to him. Shares fell about 8% after hours Wednesday. For anyone buying GPUs, keyman risk at your largest supplier just turned live: Huang has roughly 60 direct reports and the company has no obvious successor or succession plan. House Republicans say they will subpoena Huang over the alleged Chinese spy connection if he declines a closed-door briefing. TechCrunch

• OpenAI's quarterly revenue reportedly passes $10 billion. Bloomberg, citing unnamed sources, reports OpenAI's quarterly revenue crossed $10 billion for the first time, up almost 50% from $6.7 billion in the second quarter, with its adjusted operating margin improving from -150% to -109%. Its annual run rate is reportedly $50 billion. The comparison your vendor negotiations will be priced off: Anthropic reported $17 billion in Q3 revenue and an $81 billion annual run rate as of end of October — though the two count partner sales differently. TechCrunch

🌍The big AI stories today

• United Airlines — is banning third-party personal AI agents from booking, citing credential risks and possible federal-rule violations — a marker for where agent-access fights land outside tech. CBS News

• Databricks Genie — announced a second-generation always-on enterprise agent; the new mode is only coming soon to preview, priced on usage with spend limits rather than per seat. TechCrunch

• Anthropic roadshow video — revealed Claude now trains and serves on all four major chip families, including an AMD port done in three days — less compute lock-in at your main model vendor. Simon Willison

💡One takeaway