Anthropic cuts disclosure deadline to 30 days
High- and critical-severity bugs · from January 1, 2027

ON AIR
ACN · AI Coding News, the podcast. New every weekday.
Sponsor
AI models still lag at judgment calls. AI agents find a zero-day on their own. And Claude now writes ninety percent of its team's code. From AI Coding News, this is your midday AI news update for Friday, November 20th.
Our top story: the AI testing nonprofit METR has updated its time horizon graph with tasks that need judgment. METR reports some AI skills are still improving exponentially. But when models designed AI evaluations for it, none was good enough to use in its research. Why it matters: for teams handing agents open-ended work, the judgment calls still need a person in the loop, or verification gates.
Next, an AI hacking tool. Amazon says attackers used Cairn, an open-source AI agent framework, to find an unknown flaw on their own. The flaw was in MONAI Label, a medical imaging app. In about eleven minutes, the agents went from mapping it to writing files on its server. Its maintainers shipped a fix on Wednesday, November 18th. Why it matters: teams running open-source software have less time to patch it.
Don't go anywhere. Next: how Anthropic checks the code Claude writes. Your coding agents say they're done. Are they? Levr is the tracker and control plane that verifies and gates your coding agents' work against the plan. Levr holds the plans, issues, tests and gates for every agent doing the work. And work only moves forward when the gates are verified, not when the agent claims it's done. Levr. Try it free at levr.one. Welcome back.
And finally, Anthropic. It told The New Stack that Claude now writes and reviews about ninety percent of the Claude Code team's code. Cheap checks run first: the build, linters and tests. And a change can never edit the check that judges it. From June to September, merged pull requests more than doubled, while reverts fell by more than forty percent. Why it matters: teams running coding agents can copy the Claude Code team's approach: cheap checks first, model review where tests can't reach, and a person accountable for every change.
That's your midday update for Friday, November 20th. Sources are linked in the description. ACN is brought to you by Levr, the tracker and control plane that holds your coding agents' work to the plan, at levr.one.