Latest / Decrypted: The UK Cyber Briefing / Microsoft fixed a maximum-severity Azure AI flaw and told customers to do nothing
Transcript
- 0:03This is Decrypted. Saturday, the nineteenth of September.
- 0:07Microsoft fixed a maximum-severity Azure AI flaw and told customers to do nothing.
- 0:14Microsoft fixed a maximum-severity flaw in Azure AI Foundry this week.
- 0:19And its advice to customers?
- 0:21Do nothing. That's reassuring, and slightly unsettling too, because "nothing to do" also means nothing to che
- 0:29ck.
- 0:33Here's what Microsoft's advisory for C V E twenty twenty-six, eight five eight eight nine says.
- 0:40It describes "missing authentication for critical function in Azure AI Foundry".
- 0:46It says this "allows an unauthorized attacker to elevate privileges over a network".
- 0:53The Hacker News reports a C V S S score of 10.
- 0:560. That's the maximum. Microsoft credits the researcher Rémy Marot with finding it.
- 1:01Both The Hacker News and SecurityWeek report no evidence of exploitation.
- 1:07Microsoft has fixed it on its own side of the wall, so customers don't need to patch or reconfigure anything.
- 1:13Here's an analogy. Picture an office where every front door needs a keycard.
- 1:19But round the back, there's a service hatch that opens into the staff corridor.
- 1:23And nobody ever asked who was crawling through.
- 1:25"Missing authentication" means the function could be reached, and it never asked who was calling.
- 1:31Nobody had to pick a lock, because there was no lock.
- 1:35Microsoft hasn't said which function it was.
- 1:38So we don't know how much a caller could have reached.
- 1:41And also. Fixed for you, but not by you.
- 1:44The same day, SecurityWeek reports, Microsoft fixed 18 vulnerabilities across Azure and Copilot products.
- 1:52Mostly privilege escalation bugs.
- 1:54None flagged as exploited.
- 1:56All remedied server-side. The Hacker News lists others near the top of the scale.
- 2:01C V E twenty twenty-six, eight five eight eight five is a command injection in Microsoft 365 Copilot, with a
- 2:10C V S S of 9.9. C V E twenty twenty-six, eight five eight seven eight is in Azure Database for PostgreSQL, als
- 2:21o 9.9. And C V E twenty twenty-six, eight seven seven zero one is in Azure Cosmos D B, at 9.
- 2:306.
- 2:33The vendor found and fixed problems before customers had to scramble.
- 2:38But look at it from the director's chair.
- 2:40You learned about a perfect-score flaw in a platform holding your AI workloads after it was closed.
- 2:46And your only evidence that nobody walked through it is Microsoft's statement that it found none.
- 2:51I'm not suggesting that statement is wrong.
- 2:54I'm pointing out that you can't test it.
- 2:56And finally. The Secure by Design reading.
- 3:00The design decision that would have prevented this is a dull one.
- 3:03Every function authenticates its caller, every time, rather than trusting that the front door did the job.
- 3:10AI platforms are being assembled at speed.
- 3:13And authentication isn't a feature to bolt on in the next sprint.
- 3:17So what should UK organisations do?
- 3:20This is my opinion, not advice.
- 3:22First, know which AI workloads you run on the platform, and what identities, data and credentials they hold.
- 3:30Second, keep your own logs.
- 3:32And ask your supplier how long its logs go back, and what "no evidence of exploitation" rests on.
- 3:39Third, put cloud-side fixes on the risk register, even when there's nothing to patch.
- 3:45Control you can't exercise is still a dependency.
- 3:48That was Decrypted. The written version and every source are linked in the show notes.
- 3:53Thanks for listening.