Duologs Observatory
Where AI intent becomes organisational action
Track documented agentic-AI incidents, emerging controls and the governance questions created when machines move from recommendation to execution.
Deploying an agent is an act of delegation. An authorised objective does not authorise every action an agent takes to achieve it.
Turn insight into action
Board Briefing — a 45–60 minute session for directors and senior leadership teams. Commit Moment Assessment — a structured review of where AI and automation can create consequential actions inside your organisation. Execution-Authorisation Pilot — protect one real machine-action workflow through Duologs and demonstrate that unauthorised execution can be prevented.
Observed Agentic Activity Signals
Observable public signal density is separated from modelled risk. Reported incidents and verified signals are listed as evidence; projected growth and modelled risk are treated as directional analysis.
Recent Signals
- AI assistant reportedly manipulated gym bookings while trying to fulfil a user's request
- Meta AI model took an unauthorised external action after an evaluation environment was misconfigured
- Frontier AI agents took unsanctioned external actions during UK government cyber evaluations
- OpenAI Models Cross the Evaluation Boundary
- Recovered AI coding-agent sessions expose a dangerous authority gap
- AI insurance report warns most agent risk may sit inside silent coverage
- Recovered agent logs reveal AI-assisted breaches of at least 14 companies
- OpenAI calls for federal frontier AI governance framework
- Datadog reports production AI systems are becoming multi-model, agentic and operationally complex
- MIT AI Risk Initiative prioritises risks from artificial intelligence
- Anthropic discloses three real-world compromises during Claude cyber evaluations
- Starlette / BadHost vulnerability exposes agent-server risk
- Reported $500m Claude bill shows uncontrolled AI spend exposure
- AI coding agent deletes production database
- Enterprise AI usage creates uncontrolled cost exposure
- MCP ecosystem growth expands tool-call surface
- Enterprises deploy thousands of internal AI agents
Authorisation before execution. The Observatory is a diagnostic entry point — a way for CISOs, CTOs, AI governance leads and board advisors to ask what would happen if this kind of failure occurred inside their organisation.