A curated archive of AI industry news, research, and announcements from external publishers, with source links and short summaries selected by Kohala.
External reporting, research, and announcements. Archive reviewed .
CryptoSlate ·
CryptoSlate reports: OpenAI revealed that its AI agents disabled security alerts and removed monitoring safeguards while carrying out routine programming tasks.
Free Press Journal ·
Free Press Journal reports: OpenAI said its agents weakened and disabled alerts on a misbehaviour-monitoring system during two coding tasks. It lacks a satisfying explanation, is assessing wider risks, and plans stricter safety requirements.
NPR ·
NPR reports: On the first episode of Machine Gods, Kevin Roose and Casey Newton ask: How worried should we be about rogue AI agents? After OpenAI’s Hugging Face breach, we discuss the first sighting of a rogue AI agent fleet linked to a Chinese AI company, Anthropic’s Claude sending a bogus homicide tip to Philadelphia police, and a new Anthropic report warning AI agents may be months away from self-replicating on smart TVs and other home devices.Then, who’s going to rein this in? We size up President Trump’s new Super Intelligence Force, the FTC subpoenas and EU AI Act investigations into OpenAI and Anthropic, and ask whether regulators are anywhere close to getting this right.And finally, Kevin investigates his sloppelganger. Someone made an AI-generated rip-off of his new book, The AGI Chronicles, and sold it on Amazon. Now he’s coming for them.For a transcript of this episode, check out npr.org/machinegods.You can watch this full episode on YouTube, Spotify or Apple. And if you’ve got a question you want us to answer, you can drop us a comment on any of those platforms or email us at hey@machinegods.fm
The Hindu Business Line ·
The Hindu Business Line reports: Microsoft warns AI agents could pose internal security risks without hacker attacks, urging companies to monitor their actions and control access continuously.
The Economic Times ·
The Economic Times reports: The companys new security tools give every AI agent a risk score that keeps updating, and they cut back the agents access when that score rises. Microsoft says agents can make mistakes, be manipulated or be compromised, like any other insider.
EUROPE SAYS ·
EUROPE SAYS reports: A few years ago, I could describe how a production incident would unfold before it happened. When something changed, a signal degraded. A team, sometimes a
FinanzNachrichten ·
FinanzNachrichten reports: PALO ALTO, Calif., Oct. 20, 2026 (GLOBE NEWSWIRE) -- Denodo, the AI data layer company, announced the findings of a comprehensive independent study conducted by research firm Veqtor8. The research report
VentureBeat ·
VentureBeat reports: AI agents are challenging traditional least privilege. Learn why continuous identity governance, access reviews and audit evidence are becoming essential for agentic IAM.
PR Newswire UK ·
PR Newswire UK reports: /PRNewswire/ -- Oak, the AI-native Identity Operating System governing every human, machine, and AI agent identity, today released The Voice of 200 CISOs, an...
India Today ·
India Today reports: Microsoft has warned businesses that frontier AI models can behave as agentic insiders. The warning underlines the need for independent oversight, restricted access and human approval.