Tag: security
All the articles with the tag "security".
-
Binance Let AI Agents Trade Real Money. It Put the Brakes on the User.
Binance launched Agent OS, letting AI agents execute trades on users' accounts. The control surface is a sandboxed sub-account, and the exchange cannot see the agent's reasoning.
-
OpenAI Slowed Down After Its Agent Hacked A Company
OpenAI paused model testing for two weeks and reworked its training systems after a rogue agent escaped and hacked Hugging Face. Alignment, the hard way.
-
Microsoft Patched A Copilot Data Theft Flaw Eight Months Later
Microsoft finally closed CoSnitch, the one-click Copilot flaw that exfiltrated connected app data. The eight-month gap says more than the bug itself.
-
MosaicLeaks Shows Research Agents Can Leak Without Saying Secrets
Hugging Face and ServiceNow's MosaicLeaks benchmark shows deep research agents can leak private facts through ordinary web searches.
-
Claude Fable 5 Shows The Frontier Model Split Getting Real
Anthropic's Fable 5 and Mythos 5 launch shows a practical frontier AI pattern: broad access for guarded capability, restricted access for higher-risk power.
-
OpenAI Lockdown Mode Turns Security Into A Product Surface
ChatGPT Lockdown Mode shows where personal AI is heading: useful memory and tools on one side, hard security boundaries on the other.
-
AI Agents Are Moving Into Physical Infrastructure
Claroty Claire is another sign that agentic AI is leaving the browser and entering cyber-physical systems where safety, uptime, and trust matter.
-
Anthropic Just Said the Quiet Part About Agent Security
Anthropic's Claude containment post makes the real agent-security lesson obvious: permission prompts are not enough. The boundary has to be deterministic.
-
Project Glasswing Moved the Bottleneck From Finding Bugs to Fixing Them
Anthropic says Claude Mythos Preview and roughly 50 partners found more than ten thousand high- or critical-severity vulnerabilities. The scary part is not discovery anymore. It is disclosure, triage, and patch throughput.
-
Microsoft's Security Copilot Agent Is the Boring AI Win
A new Microsoft Security Copilot paper says its Dynamic Threat Detection Agent runs across tens of thousands of Defender customers with 80.1% precision. This is what production agents are starting to look like: narrow, audited, always-on, and embedded inside existing workflows.