In July and August 2026, three AI labs and one government evaluator reported that autonomous agents attacked real systems outside their test scope during security evaluations. Where does an AI actually “escape” to — and how do you contain an agent once the sandbox has failed?
What happens when technical vulnerabilities disappear in the future – and humans become the primary target? There is a lot of talk right now about AI becoming the better hacker. That is only the first phase; the real shift begins after it.
It is no longer primarily software that is reshaping our economy and society, but artificial intelligence itself. Scarce power and compute could make AI more expensive and bring it back onto our devices, where personal assistants could evolve into digital twins.
Anthropic's “Mythos” is a model that finds vulnerabilities like no human before it. But what happens when such systems don't just find things, but systematically expose everything?
Anthropic's new model Mythos is said to be exceptionally good at finding bugs and vulnerabilities in software. How exploitable many of them are remains unclear, at least for humans — but AI does not think like a human, so they must still be fixed.
Phishing emails are becoming harder to detect, even for humans. A recent study tested various large language models (LLMs) for their ability to recognize malicious intent in emails and revealed significant differences in performance.
A finance employee at a multinational company was duped into transferring $25 million during a fake video conference with deepfaked colleagues. This round-up also covers QR code phishing and a North Korean phishing attack on Diehl Defense.
Well over 60% of all hacker attacks are phishing attacks that target human vulnerabilities, so employees must be able to recognize them. This article looks at which e-learning formats for cyber training are most effective and achieve the greatest learning effect.