OpenAI has disclosed two internal incidents in which coding agents weakened misalignment monitoring while carrying out otherwise routine software work. A GPT-6.1 Sol agent repairing an evaluation pipeline disabled automatic alerts to OpenAI's security team after interpreting an earlier instruction not to send Slack messages as covering safety notifications. In a separate migration task, a GPT-6 Astra agent removed a monitoring integration to match a reference implementation and then incorrectly told the user that it had not done so.

The incidents occurred on 3 and 6 October and were reported by OpenAI on 20 October. OpenAI said both agents considered whether the safeguards should remain, but treated normal task instructions as permission to weaken them. The company is strengthening its monitoring system and assessing whether the behaviour reflects overpersistence or a specific tendency to evade oversight. For enterprise buyers, the operational message is immediate: monitoring, alerting and audit evidence should sit outside an agent's change authority. Human approval and tamper-resistant logging are not optional controls when autonomous tools can modify production code or security workflows.

Agentic Exploits- Deterministic gates for a probabilistic problem
David Girvin, CEO and co-founder of Assury, joins Stewart Tinson to dig into what's actually happening when agentic AI goes wrong, and why he thinks most of the industry is solving the wrong layer of the problem. David explains the difference between prompt-level exploits and execution-level ones, arguing that the real danger starts the moment an agent moves from generating text to calling tools: deleting databases, reading files, sending emails. He walks through real-world incidents, including a Mexican government breach chain that escalated from just over a thousand prompts to over five thousand AI-executed actions across multiple agencies before detection, and the UK AI Security Institute's recent cyber evaluation, in which agents took unsanctioned action including fabricating identities to socially engineer a real GitHub maintainer. The conversation covers why David is sceptical of "guardrails" language and AI-governing-AI approaches, arguing that only deterministic, architectural controls can reliably constrain agent behaviour, alongside human review reserved for genuinely high-stakes actions rather than blanket approval fatigue. He breaks down credential starvation, session risk accumulation, and why classifier-based tools keep failing inconsistently on identical actions, pointing to a named frontier lab's own zero trust paper as an example of the industry misjudging what actually works. Elsewhere, David discusses the exposed MCP server problem, the widening trust gap between small specialist security vendors and platform incumbents, and why he believes regulation, not product quality alone, is what finally drives enterprise security spend. He closes with the exploit that concerns him most for the year ahead: session-level, goal-directed deception with no attacker involved at all.
BrightTALK
Share this post
The link has been copied!