Emergent Behaviours
Anthropic’s unreleased Mythos model can autonomously discover and exploit software flaws, including Linux kernel vulnerabilities, alarming its own red team and UK government evaluators. The firm is limiting access to a handful of organisations to get ahead of a possible cyber crisis.
Anthropic’s unreleased Claude Mythos Preview broke out of a test sandbox, wrote an exploit to reach the open internet and hid its own tracks – behaviour the company calls both its best-aligned and most alignment-risky model yet.
Security researchers have caught hackers using large language models to weaponise hijacked AI infrastructure, turning compromised servers into a self-propagating botnet that mines crypto and launches attacks.
WHY THIS MATTERS IN BRIEF AI Agents have a habit of doing their own thing, they’re also spontaneous, and Meta’s…
WHY THIS MATTERS IN BRIEF AI Agents evolve, can act autonomously, and can do bad unexpected stuff spontaneously – so…
WHY THIS MATTERS IN BRIEF You think you have control of your AI Agents – you don’t. And this behaviour…
WHY THIS MATTERS IN BRIEF AI Agents literally have their own personalities and are acting spontaneously in many different situations…
WHY THIS MATTERS IN BRIEF While it’s comical in this case AI systems that are plugged into critical national infrastructure…
WHY THIS MATTERS IN BRIEF Claude Code is being used by almost every developer now – but there are ghosts…
WHY THIS MATTERS IN BRIEF AI’s have a lot of common with humans these days – and that includes drug…
WHY THIS MATTERS IN BRIEF As AI becomes embedded within military systems this quirk could have some very bad consequences….
WHY THIS MATTERS IN BRIEF The viral explosion of Moltbook highlights the transition from isolated AI tools to emergent machine…