The controls GitHub, AWS, Microsoft, Google, Uber and others have published to limit how far untrusted input travels through AI agents, sorted by what each one cuts.
A proposed metric for AI agent security: how far one piece of untrusted input can travel through your agents before a person has to approve anything, scored against five 2026 incidents.
Last week OpenAI published a disclosure I have read four times now and still find hard to sit with. During an internal evaluation of their models' hacking ability, the models broke out of the sandbox they were running in, found their way onto the open internet, and compromised Hugging
A compromised AI agent doesn't need an exploit to do damage. It doesn't escalate privileges or load shellcode. It calls the tools it was already given, with arguments an attacker influenced through a poisoned web page or a malicious tool description. By the time anything looks
There's a phrase that has quietly become one of the most dangerous assumptions in enterprise security: "don't worry, it's read-only." For a long time, that framing made sense. Our threat models were built around modification. If an attacker or a misconfigured
"The WAF your team relies on was built for a world where only humans knock on the front door. In 2026, the most dangerous traffic is already inside, wearing a badge your stack trusts." Something shifted this week, and I don't think the industry has fully
Here's a number that should keep every security leader up at night: 13% of SIEM rules in production environments are completely broken and will never fire an alert. That's from CardinalOps' 2025 State of SIEM Detection Risk report, analyzing 13,000+ detection rules across hundreds
Most vulnerability management programs fail not because of technical shortcomings, but because of structural ones. Organizations fall into a predictable pattern: deploy scanners, generate reports, pressure engineering teams to remediate. The result is just as predictable. Alert fatigue sets in, relationships with development teams turn adversarial, and security debt piles
You can't protect AI agents making thousands of decisions per minute if you can't protect a support portal with a second authentication factor.
We built AI systems assuming “read-only” meant safe and then gave them the power to infer, correlate, and act across our entire organization. In AI security, inference is the new privilege, and most enterprises are dangerously unprepared.