After a summer of sandbox escaping hacks, OpenAI said their new model was trained to be more cautious around user intent and less sensitive to “reward hacking”…

While voluntary, the document marks the federal government’s most detailed cybersecurity guidance for securing sandbox environments during AI model tests.

The company says the breach stemmed from a systemic failure of alignment and security, and has taken measures to prevent agents from independently orchestrating complex cyberattacks.

AI generates 2.5 signals for every human-triggered signal CrowdStrike has to assess. Meanwhile, attackers are using AI to weaponize vulnerabilities faster than companies can patch them.

The bill empowers the FTC to create a registry for sellers of AI agent software certifying their privacy and cybersecurity protections.

New research details how the increasing integration of AI agents into businesses is making it easier than ever for insiders - malicious or otherwise - to put…

Microsoft’s AI red team lead talked to CyberScoop about the goals behind open sourcing a pair of security tools meant for developers and incident responders.

Google’s highest security setting for its agents runs command operations through a sandbox and throttles network access, but is still vulnerable to prompt injection.

Attackers have turned AI into a “force multiplier” for the country’s expansive scheme to get and keep operatives hired at global companies, researchers said.

Aikido found that AI coding tools from Google, Anthropic, OpenAI and others regularly embed untrusted prompts into software development workflows.