AI security concerns span cyberattacks, privacy and weapons
AI agents’ expanding capabilities are raising security concerns as they interact with apps, access documents and perform tasks for users. Apple is restricting AI apps’ access to Mac users’ private data. Google reports that state-backed groups linked to Russia, China, Iran and North Korea are chaining AI agents to discover vulnerabilities and conduct intrusions without human intervention.
The European Commission says it has contained most of a breach involving the public tool ARTEX and is investigating whether employee personal data was compromised. North Korean group BlueNoroff is using commercial AI tools to impersonate executives through deepfake faces and voices. OpenAI has patched a ChatGPT vulnerability that enabled malicious code execution through shared chats, while hidden, persistent instructions pose another threat to enterprise assistants.
Concerns also extend to oversight and military use. The European Central Bank has flagged financial stability risks, including funding challenges and pressure from cheaper models. The UN warns that AI is taking over human decisions in warfare. Anthropic says it blocked accounts linked to Houthi efforts to develop guided missiles and simulate hypersonic weapons, while acknowledging that its safeguards missed some requests.