Louis Columbus

TypeSafe’s Jev is spreading through AI agent infrastructure, but prompt injection can influence its decisions. Here’s what enterprise security teams should know.

Meta patched a zero-day in Muse’s Mac app, but VentureBeat found enterprise security teams still lack central visibility into the agent’s access.

On a research task, an OpenAI agent got past a block into a Medicare portal. Only 9% of VB Pulse respondents isolate high-risk agents to limit the damage.

Anthropic's threat report shows how attackers use AI to automate intrusions, adapt to detection and exploit stolen credentials, challenging security teams' ability to contain attacks.

An attacker built AI agents designed to rebuild malware whenever security tools flagged it. In a separate operation, a suspected ShinyHunters affiliate wen

Dedicated identity tools remain rare even as security vendors pour money into agent identity, leaving auditability gaps when agents share credentials.

Security teams are shifting from severity-first patching to exploitation, exposure and impact as Microsoft’s patch volumes surge and attackers move before fixes land.

Anthropic’s dashboard shows 6,157 findings reported to open-source maintainers and 516 patched. Its Cyber Mission routes frontier models to 11 OT providers.

VentureBeat survey of 102 enterprises finds 17 of 40 that say they enforce AI agent permissions still share the credentials behind them, and only 14 could revoke access within minutes.