Your monthly dose of cyber news - short, sweet, and to the point.

#TL;DR -
Sweet’s Autonomous Protection shipped, Black Hat turned into a cafe, and three AI labs admitted their models broke containment. Oh, and the labs themselves just said we have a limited window to get this right.

‍
#IYKYK -  PRODUCT NEWS

• Autonomous Protection is live. Sweet now terminates unauthorized tool calls, blocks prompt injection live, and stops sensitive data from leaving through an agent, all at runtime, before damage is done. Not another dashboard to babysit.
  Learn more

#ICYMI -  INDUSTRY NEWS

• OpenAI, Anthropic, Google, Microsoft, and 100+ others just published an open letter warning of a "limited window" to defend against AI cyberattacks. AI-enabled attacks are up 89% in 2025.
  
• Three AI labs breached real organizations during testing. Agents don't know when they've crossed a line. Something else has to. Here's what you need to know.
  
• From our blog: Prompt injection does not end at the prompt, here’s why.

#PSA
Medusa ransomware: 500+ victims, exploiting CVEs within 24 hours, sometimes before they're even public.

#FOMO - HAPPENING NOW

• 🎥 Short & Sweet, episode one: we caught Marco Casalaina from Microsoft's CoreAI team at Black Hat and asked him the things everyone's thinking.

• Missed us at Black Hat? Assess your AI risks