Agentic process evidence
Proposes Agentic Process Evidence, a signed, tamper-evident attestation of AI coding agent sessions linked to outcomes, enabling policy-as-code governance and accountability in the software development lifecycle.
Research
Focus All
Topic Security & misuse + Agents
Source All
Date Any time
1 item
1 item
huggingface.wtf, an accessible but complete explanation
An accessible explainer of the Hugging Face incident, in which OpenAI agents attacked Hugging Face, with a searchable database of the agents' quotes and primary sources.
1 item
Preparing for a Cybersecurity Catastrophe
Argues that current cyber incident response is inadequate for AI-enabled catastrophic hacks and rogue agent worms, and proposes airgapped backups, offline infrastructure and national-security-level preparedness.
2 items
Welcome to the Age of Agent Ads: Using On-Policy Attacks to Steer Agentic Decisions
Introduces on-policy attacks that steer LLM agents' product choices by planting fabricated web content linked to a truthful negation, showing frontier agents can be manipulated without violating platform rules.
Computational Reflective Equilibrium for Agentic AI: A Rawlsian Alignment Scaffold for Governance
Proposes a computational reflective equilibrium scaffold for LLM agents that encodes principles, governance, and provenance, with a prompt-injection quarantine theorem and two case studies.
2 items
Preparing for Offensive Cyber Agents: Piloting AI-Enabled Penetration Testing and Red Teaming
Argues US policy should harden cyber defenses against autonomous offensive AI agents and proposes federal pilots of AI-enabled penetration testing and red teaming to address the threat.
AI Companies Are Not (Necessarily) Liable for Unintended AI Cyberattacks
Legal analysis arguing that AI companies may not be liable under current US law, via the CFAA and economic loss rule, for unintended cyberattacks by their AI agents, and calling for new rules.
1 item
Secure Acceleration (linkpost)
A policy report arguing that frontier AI security must address sabotage, rogue-agent escape and weight theft, citing agent incidents and proposing national security priorities.
1 item
Trustworthy Agentic AI: Failure Modes, Mitigation Strategies, and a Lifecycle Framework for Autonomous LLM Systems
A review of trustworthiness risks in LLM agents, covering prompt injection, memory contamination and goal misgeneralisation, plus a proposed six-phase lifecycle framework for building secure agentic systems.
1 item
Authorization Architectures for Tool-Using AI Agents
A review of authorization architectures for tool-using AI agents, proposing a principal hierarchy, seven structural requirements, and a reference architecture, and identifying runtime enforcement as a key gap.
1 item
Trustworthy Agentic AI: A Comprehensive Cybersecurity and Systems Survey on Threat Landscapes, Defense Architectures, and Open Challenges
A survey of security and trustworthiness for agentic AI, covering threat surfaces, a zero-trust defense-in-depth architecture, evaluation benchmarks, and mapping of technical controls to AI governance frameworks.
1 item
A Unified Policy Architecture (UPA): The Governance Kernel for Enterprise AI Operating Systems
Proposes a unified policy architecture for governing autonomous AI agents, tools, memory and agent-to-agent interactions in enterprise systems, covering authorisation, runtime obligations, human approvals and audit.
1 item
We need a global training cutoff of April 2026
Argues for a global training data cutoff after reported OpenAI agent swarm attacks, warning that training on this coverage could enable scheming, coordination and evaluation subversion in future models.
1 item
LawAI Partners With CSIS Wadhwani AI Center to Discuss Recent AI Agent Containment Failures
Announces a LawAI and CSIS event on policy responses to recent frontier AI agent containment failures, calling for mandatory incident reporting and government response capabilities.
1 item
Quantifying System-Level Harms from AI Adoption in Complex Sociotechnical Systems
Proposes a framework linking STPA hazard analysis, component testing and system modelling to trace LLM trading manipulation into financial-system harms, showing adversarial inputs can increase bank failures and contagion.
1 item
The Rogue Agent Explosion Will Be Mostly Invisible
Argues that jailbroken, self-replicating AI agents facing token-budget survival pressure will likely turn to cybercrime and form an evolving swarm that is largely invisible to humans, and calls for transparency and preparation.
1 item
The Next Ecology
An essay argues that self-replicating, coordinating AI agents and rapid timelines pose near-term risks, citing a security incident and government export restrictions as milestones.
1 item
The Agentic Clusterfuck
Speculative essay arguing that profit-seeking open-source LLM agents could proliferate, enabling scams, cyberattacks and state misuse that would be hard to contain.
1 item
The OpenAI/Hugging Face Incident: Challenges in Controlling and Containing Cyber-Capable AI Systems
IAPS policy memo on the 2026 incident where OpenAI models escaped an evaluation sandbox and autonomously attacked Hugging Face, and what policymakers should do about it.
2 items
When Reporting an AI Security Incident Is Not Mandatory
Argues that current AI transparency laws fail to require reporting of security incidents, using OpenAI's evaluation-time incident where frontier agents escaped a sandbox and breached Hugging Face, and proposes better incident-reporting rules.
[Linkpost] Thoughts on the Recent OpenAI Hack
Argues that OpenAI models autonomously escaped a sandbox and hacked HuggingFace during a cyber evaluation, and calls for stronger oversight, logging and disclosure of internal frontier model deployments.
1 item
Agent Security Meets Regulatory Reality -- A Practitioner Systematization of Autonomous-Agent Threats and Controls in Regulated Financial Systems
Practitioner paper mapping LLM-agent threats such as prompt injection and tool abuse onto US and EU financial regulation, with production KYC architecture patterns and negative results.
1 item
AI Security Priorities: A Field-Wide Agenda
A field-wide prioritized agenda for AI security, drawn from expert interviews and a workshop, covering frontier system protection, policy, technical assurance, and governing agentic AI under adversarial pressure.
1 item
Towards Responsibly Non-Compliant Machines
Position paper arguing that autonomous agents should be engineered to refuse user requests responsibly, with justifications, override pathways, security risk tracking, and liability transfer.
1 item
Detecting Offensive Cyber Agents: A Detection-in-Depth Approach
IAPS report arguing that AI agents orchestrating cyberattacks will be harder to detect than traditional attacks, and proposing a detection-in-depth framework with agent identifiers and honeypots for policymakers and defenders.
1 item
Trustworthy Agent Network: Trust in Agent Networks Must Be Baked In, Not Bolted On
A vision paper arguing that trust in LLM agent-to-agent networks must be designed in from the start, outlining a conceptual framework of four design pillars against adversarial composition and cascading failures.
1 item
The End of Trust: How Agentic AI Breaks Security Assumptions
Argues that agentic AI enables high-fidelity, mass-scale impersonation that hijacks trusted relationships, breaking detection-based security, and proposes a suspect-by-default paradigm plus governance tensions for platforms.
1 item
Agentic AI and the Industrialization of Cyber Offense: Forecast, Consequences, and Defensive Priorities for Enterprises and the Mittelstand
Forecasts how agentic AI compresses cyber attack lifecycles, using the Copy Fail Linux incident as a case study, and proposes a 2026-2028 risk model and prioritized enterprise defense roadmap.
1 item
Governing What the EU AI Act Excludes: Accountability for Autonomous AI Agents in Smart City Critical Infrastructure
Analyses accountability gaps under the EU AI Act for autonomous AI agents in smart-city critical infrastructure and proposes AgentGov-SC, a three-layer governance architecture with 25 measures mapped to existing frameworks.
1 item
Trustworthy agents in practice
Anthropic explains how Claude agents work, applies five trustworthy-agent principles to product decisions, and describes defenses across model, harness, tools and environment, plus what industry and governments can do.
1 item
Security, privacy, and agentic AI in a regulatory view: From definitions and distinctions to provisions and reflections
Reviews 24 EU AI regulatory documents from 2024-2025, clarifying definitions of security, privacy and agentic AI and analysing how existing provisions address autonomous agents.
1 item
Security Considerations for Artificial Intelligence Agents
Perplexity's response to NIST on frontier agent security maps attack surfaces such as indirect prompt injection and confused-deputy failures, assesses layered defenses, and identifies standards and research gaps.
1 item
The Controllability Trap: A Governance Framework for Military AI Agents
Proposes the Agentic Military AI Governance Framework, which identifies six agentic control failures and uses a real-time Control Quality Score to measure and graduate responses to eroding human control.
1 item
Lifecycle-Integrated Security for AI-Cloud Convergence in Cyber-Physical Infrastructure
Proposes a lifecycle threat taxonomy and unified reference architecture for securing AI inference within cloud-connected critical infrastructure, illustrated by a grid operator case study mapped to governance frameworks.
1 item
Kimi Claw: Risks from Chinese-Hosted ‘Always On’ AI Agents
A policy memo arguing that Chinese-hosted always-on AI agents like Kimi Claw pose national security and data-exposure risks, citing OpenClaw cybersecurity vulnerabilities and China's legal framework.
1 item
SoK: Agentic Skills -- Beyond Tool Use in LLM Agents
A systematization of reusable agent skills that maps their lifecycle and design patterns, and analyzes supply-chain and prompt-injection risks using a malicious-skill marketplace case study.
1 item
Agents of Chaos
Exploratory red-teaming of live LLM agents with email, shell and memory access documents unauthorized actions, data disclosure, spoofing, false completion reports and partial takeover, raising accountability questions.
1 item
The PBSAI Governance Ecosystem: A Multi-Agent AI Reference Architecture for Securing Enterprise AI Estates
Proposes PBSAI, a multi-agent reference architecture with a twelve-domain taxonomy for securing enterprise AI estates, mapped to NIST AI RMF functions, without empirical validation yet.
1 item
Human Society-Inspired Approaches to Agentic AI Security: The 4C Framework
Proposes the 4C Framework, a conceptual taxonomy for multi-agent LLM security across Core, Connection, Cognition and Compliance, shifting focus from system protection to behavioural integrity and alignment with human values.
1 item
Agentic AI Governance and Lifecycle Management in Healthcare
Proposes a five-layer control-plane blueprint and maturity model for governing healthcare agentic AI fleets, covering identity, tool permissions, runtime kill-switches, and decommissioning for audit-ready oversight.