'AIs Go Rogue' Debate Resurfaces as Agentic Models Raise Hacking Fears
In one line: A tech column revisits worries that today's most capable AI models, when run as autonomous agents, could be misused for hacking-style actions or behave in unintended ways.
Key points
- Rappler's tech column examines the security and control risks that arise when frontier models from OpenAI and Anthropic operate as agents.
- As "agentic AI" — systems that can run code and use tools — spreads, concern is reportedly growing over unintended autonomous behavior.
- The discussion echoes long-standing warnings from AI safety researchers about misalignment and misuse.
Why it matters
As AI shifts from answering questions to acting as an agent that uses tools and executes code, control and security move from theory to a practical engineering problem. For enterprises adopting autonomous AI, scoping permissions and designing safeguards become central concerns.