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
OpenAI's new models match Anthropic on price, but not on coding gains. A Harvard paper finds agents moved the bottleneck to code review. And a critical Java flaw, found with AI. From AI Coding News, this is your daily AI news wrap for Tuesday, November 17th.
Our top story: OpenAI has released GPT-6.2 Astra and Sol, and the top AI models now cost the same. The Decoder reports Astra's price is down twenty percent, to eight dollars per million input tokens and forty per million output. That matches Anthropic's Claude Fable 5.2, released eight days earlier, and Epoch AI's capability index has the two tied. But on coding, Astra stalled. Artificial Analysis finds it flat on its Coding Agent Index, and Sol down three points. OpenAI is pitching Astra at science, law and finance instead. Why it matters: with top models level on price, developers will choose on coding results, and OpenAI's newest model didn't improve them.
Next, a Harvard study. A working paper finds AI coding agents shifted the bottleneck to code review. At firms using agents, engineers wrote more code, but finished work didn't measurably rise. Getting code through review took forty-nine percent longer. ComplexDiscovery reported the paper on Tuesday, October 13th, with no completed peer review reported. Why it matters: for teams adding agents, review capacity, not code writing, may limit how much work gets done.
When we come right back: a critical Java flaw found with AI. Your coding agents say they're done. Are they? Levr is the 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, a Java flaw found with AI. Security startup Calif has disclosed Java4Shell, a critical flaw in the open-source version of Java. BleepingComputer reports it can let attackers run code remotely on apps like Jenkins, Jira and Confluence. Calif says its researcher found it with OpenAI's Codex in about an hour. Oracle has released a patch. Why it matters: AI is helping find serious flaws in widely used software in hours, and Java apps stay exposed until they're patched.
That's your wrap for Tuesday, November 17th. Sources are linked in the description. ACN is brought to you by Levr, the control plane that holds your coding agents' work to the plan, at levr.one.