Safety
Import AI 453: Breaking AI agents; MirrorCode; and ten views on gradual disempowerment
Import AI issue 453 covers research and developments around vulnerabilities in AI agent systems, including methods for breaking or adversarially manipulating AI agents. The issue also features MirrorC
Import AI issue 453 covers research and developments around vulnerabilities in AI agent systems, including methods for breaking or adversarially manipulating AI agents. The issue also features MirrorCode, likely a tool or technique related to code generation or AI-assisted programming, and presents ten distinct perspectives on the concept of gradual human disempowerment as AI systems become more capable and autonomous.
Related
- OpenKedge: Governing Agentic Mutation with Execution-Bound Safety and Evidence Chains
- AgentCity: Constitutional Governance for Autonomous Agent Economies via Separation of Power
- Governance-Aware Agent Telemetry for Closed-Loop Enforcement in Multi-Agent AI Systems
- Scheming in the wild: detecting real-world AI scheming incidents with open-source intelligence
- Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines
- WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks
Source: Import AI | 2026-04-13