In today’s 2-Minute Tech Briefing, OpenAI’s new “confession” system that flags AI mistakes, Microsoft’s quiet fix for long-abused Windows shortcut vulnerabilities, and why CIOs are pulling back from AI hype as sales quotas fall across major vendors.
Register Now
Hello and welcome to your 2-Minute Tech Briefing from Computerworld. I'm your host, Arnold Davick, reporting from the floor of the New York Stock Exchange. Here are your top IT news stories you need to know for Tuesday, December 9. Up first, from ComputerWorld.
Open AI has developed a new system that trains its GPT-5 model to confess when it cheats, cuts corners or fails to follow instructions, the model now provides a second output that reveals hallucinations instruction violations or uncertainty. It's a step towards more accountable AI.
Open AI says it's proof of concept and not yet a production feature. From CSO, a long standing Windows link shortcut exploit may finally be addressed. Attackers have used hidden commands inside shortcut files for years.
Microsoft says it's not a vulnerability, but new updates now force full visibility of hidden arguments. Third-party firm 0patch says its own patch offers an additional, more aggressive fix. And finally, from CIO, AI sales expectations at Microsoft and open AI are dropping as CIOs slow their spending.
Microsoft recently reduced AI quotas. Open AI also cut projected agent revenue by $26 billion. Analysts call this a healthy correction as enterprises evaluate real world value before making long term commitments. That's today's 2-Minute Tech Briefing, for more enterprise tech news, visit Computerworld, CIO, and CSO online.
And don't forget to like and subscribe to TechTalk on YouTube.
Sponsored Links