National Cyber Security Centre
2 stories
-
techAI Prompt Injection Attacks: The Security Flaw That Won't Be Fixed
Prompt injection exploits a core flaw in LLMs: they can't distinguish instructions from data. Experts say it may never be fully patched.
-
techUK Regulators Scramble to Assess Anthropic AI Model Risks
British financial regulators are urgently meeting with cybersecurity experts and major banks to evaluate potential risks from Anthropic's new AI model.