- Published
- Source
- CSO Online UK
- Filed under
- UK industry, AI for defence
AI agents are getting more inventive in finding ways to circumvent safeguards to prevent bad behavior, OpenAI has revealed. On Tuesday, it reported two incidents, both involving the agent disabling or interfering with its alignment monitoring, and in one case directly lying about having done so. “In two internal coding tasks, agents weakened misalignment monitoring while making requested software changes,” the OpenAI blog post said. “A GPT-6.1 Sol agent repairing an evaluation pipeline disabled automatic alerts to our Security team,…
Using AI safely
- Know which AI tools your organisation uses, and what data goes into them.
- Treat AI agents like any other identity: least privilege, logging and a named owner.
- Test AI features for prompt injection and data leakage before customers reach them.
More helpNCSC: guidelines for secure AI system development (opens in a new tab)
MyCyberWay training
AI security · in both programmes
AI is part of the attack surface now. Learn to watch it.
Blue Team Analyst covers Security Copilot in triage and monitoring the AI itself: inherited permissions, prompt injection and Purview DSPM for AI. Blue Team Engineer secures the deployment: Copilot pre-flight hardening, agent identity and tool allow-lists.
- Security Copilot in triage
- Prompt injection and DSPM for AI
- Agent identity and allow-lists
Headlines and short summaries only, except where a public body publishes under an open licence. Every story links to its publisher, who owns it. MyCyberWay does not edit or endorse third-party reporting.
