ON AIR
Listen to the brief
ACN · AI Coding News, the podcast. New every weekday.

ON AIR
ACN · AI Coding News, the podcast. New every weekday.
Sponsor
Microsoft responds to its toolkit hijack. OpenAI's agents finish nearly half of longer tasks. Arizona opens a probe as banks push back on AI agents. And an AI assisted attack hits the European Commission. From AI Coding News, this is your morning AI news brief for Thursday, November 19th.
Our top story: on Wednesday, November 18th, we reported, citing BleepingComputer, that poisoned versions of Microsoft's Agent Framework carried a credential stealer. Now Microsoft tells The Register it has made some repositories private while it investigates. The Register reports apparent unauthorized access to a Microsoft GitHub admin account. So if either bad version ran on your machines, assume their cloud keys are stolen, and BleepingComputer advises rotating them right away and rebuilding from the clean 1.26.0 release.
Next, OpenAI. The Decoder reports new OpenAI data on its research agents. OpenAI says it is tracking signs of AI speeding up AI development. Its solo agents now finish an estimated 46 percent of two to four hour tasks, double March's rate. That puts them near a coin flip on multi-hour work. OpenAI cautions that its data may not extrapolate in a straight line. For teams handing agents multi-hour jobs, that means planning to review about half of the results.
Next, AI agents and banks. Arizona Attorney General Kris Mayes says her office has opened a criminal investigation into OpenAI and other AI companies, over what she calls the apparent infiltration of banks by rogue AI agents. Reuters notes an investigation does not necessarily mean charges, and OpenAI did not immediately respond. Reuters reported last month that major banks are blocking outside AI agents and flagging transactions that use logins customers shared with them. Security experts quoted by eSecurityPlanet say banks should replace shared passwords with API access that has limited permissions. If your agent touches a bank account, build it on the bank's own limited access, never a customer's password.
When we come right back: an AI assisted attack on the European Commission. 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, the European Commission. The Commission confirms a cyber campaign. The Register says systems were disrupted for over 10 days. Politico's sources say hackers used ARTEX, a public AI pen testing tool. A security CEO tells The Register such attacks now take hours, not weeks. So what does that mean for developers? A missed patch can now mean a break-in the same day, so Qualys's head of threat research says to patch internet-facing systems within hours, not weeks.
That's your morning brief for Thursday, November 19th. 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.