Freezers, a breach and poisoned training data: the attacks keep coming
The pattern of incidents is not slowing down. The Guardian reports that supermarket freezers across Europe were hit by a second suspected cyber-attack. It got 395 points on Hacker News. Our sources do not say whether AI was involved, so for now it is a security story, not an AI story.
WorkOS, a company that sells login tools to software companies, reported a security breach, with few details so far. A new paper, 'Agentic backdoors via pretraining poisoning', looks at hiding bad behavior inside the data used to train a model. It got 97 points.
On defense, OpenAI is raising its bug bounty rewards, with new bonuses for report quality, loyalty and priority campaigns. It is also preparing an internal security sprint with Cathedral Systems, using AI to test its own defenses. This comes two days after we saw researchers take over ChatGPT accounts.
Check your agents and tools today: what can they access, and who updated them last? Bigger bounties are good news, but they only help if disclosure is fast, and too often we learn about incidents weeks or months later.
