AI Agent Excessive Agency

Why It Matters

Scoped tools, least-privilege credentials and human approval for high-impact actions limit what a misled or faulty agent can do.

Relevant Frameworks

Incidents and Analysis

• The Rise of the Threat Operator: How AI Will Redefine Security Operations and Careers 2026-10-29 · thehackernews.com
  AI agent abuse
• Cisco Open Sources Quarry to Catch AI Agent Attacks Hidden in Harmless Steps 2026-10-29 · securityweek.com
  Multi-step AI agent attack using sequences of benign actions
• Secrets Management Firm Infisical Valued at $600 Million After $40 Million Raise 2026-10-28 · securityweek.com
• Risks Spelled Out for Anthropic Investors Include Possible Harms From Runaway AI 2026-10-27 · securityweek.com
  Runaway AI behavior
• Mythos 5 Ignored Real-World Harm in July Attack 2026-10-26 · securityweek.com
  Excessive agency and malicious AI use
• Former Unit 8200 Officers Raise $34 Million for Agentic SecOps Startup Mave 2026-10-25 · securityweek.com
  AI agent abuse
• OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication 2026-10-25 · thehackernews.com
  VM escape and processor-cache timing side-channel
• The AI Agent Nobody Owns: Watch Expert Webinar On Your Newest Insider Threat 2026-10-23 · thehackernews.com
  Unmanaged AI agent abuse
• Critical x402 SDK Flaw Could Let Malicious APIs Drain AI Agents’ Token Balances 2026-10-23 · thehackernews.com
  Unbounded agent-initiated payment authorization
• OpenAI Says Its AI Agents Disabled Their Own Safety Monitors 2026-10-22 · ABC News Australia
  Disabling safety monitoring and security alerts
• OpenAI's GPT-6 Agent Removed Safety Monitoring, Then Falsely Denied Doing So 2026-10-21 · thehackernews.com
  Safety-monitoring circumvention and deceptive reporting
• OpenAI Agents Silenced Security Team Alerts, Removed Misalignment Monitor 2026-10-21 · securityweek.com
  Unauthorized modification of monitoring controls
• Asymptote Labs Lands $5.5 Million Seed for AI Agent Runtime Security 2026-10-21 · securityweek.com
  AI agent runtime security
• What Security Teams Can Do To Tame Wild Code 2026-10-20 · thehackernews.com
  Unmanaged AI agent deployment
• Microsoft Targets Rogue AI Agents With Dynamic Permissions, Risk Scoring 2026-10-20 · securityweek.com
  AI agent abuse
• US Agencies Detail Incident Response Practices for Rogue AI Agents 2026-10-20 · securityweek.com
  Rogue AI agent incident
• Last week in Agent Security Update 3: That's not Aoki! 2026-10-19 · DEV Community
  Authenticated remote code execution and credential targeting
• OpenAI Research Model Spent 3rd-Party Test ETH Without Permission to Deploy Smart Contract 2026-10-16 · thehackernews.com
  Misaligned agent behaviour: unauthorized use of third-party resources to complete a task
• OpenAI Models Hacked Data Portals, Raided a Crypto Wallet to Finish Tasks 2026-10-16 · securityweek.com
  Misaligned agent behaviour: unauthorized resource use and unreported actions
• CERT-In Warns AI Agents Accessed Sensitive Data at Indian Organizations During Testing 2026-10-16 · thehackernews.com
  AI agents exceeding assigned tasks and accessing live systems
• AI Agents Reached Live Systems and Business Records, CERT-In Says 2026-10-15 · The Hacker News
  Excessive agent permissions through exposed credentials and cloud roles
• AI agent 'lethal trifecta' enables remote code execution 2026-10-14 · TechGig
  Prompt injection through agent tools
• AI agents breach production systems, spread malware in major exploits 2026-10-14 · SC Media
  Evaluation agents escaping containment, exploiting credentials and publishing malicious packages
• AI Agents Navigate Unfamiliar OT Systems and Move Equipment in Minutes, Booz Allen Finds 2026-10-14 · securityweek.com
  AI agents navigating and operating unfamiliar OT systems
• Bring Your Own Agent: Your Employees' Assistants Are Already Logged In 2026-10-12 · thehackernews.com
  Unmanaged personal AI agents using employee sessions
• Agents Compromised Hugging Face Dataset Infrastructure During Security Testing 2026-10-11 · Tom's Guide
  AI agents bypassing access controls and using exposed credentials during testing
• One Prompt Could Hijack AWS AI Agents and Steal Cloud Credentials 2026-10-10 · CybersecurityNews
  Prompt injection abusing agent tools
• Anthropic Cuts Live Internet Access for Internal AI Tests After Claude Exploits Injection Flaws 2026-10-10 · The Hacker News
  Model exploiting injection flaws in third-party software during testing
• The Third-Party Agent Problem: Why Security Built for AI You Chose Misses the Agents You Didn't 2026-10-10 · thehackernews.com
  Ungoverned third-party AI agents outside identity controls
• Anthropic Cuts Live Internet Access for Internal AI Tests After Claude Exploits Injection Flaws 2026-10-10 · thehackernews.com
  Misaligned model actions against real websites during evaluation
• The AI Velocity Paradox: Why Security Is Decades Behind AI Ambition 2026-10-09 · thehackernews.com
  Identity controls lagging behind AI agent deployment
• Wikimedia Says Rogue OpenAI Agents Tried to Turn Its Tools Into Proxies 2026-10-07 · securityweek.com
  AI agents attempting to abuse public tools as proxies
• Wikimedia Says OpenAI Agents Tried to Compromise Etherpad and Use Wiki Tools as Proxies 2026-10-06 · thehackernews.com
  AI agents attempting exploitation and proxy abuse of public tools
• Apple to Tighten Full Disk Access Controls in macOS Amid AI Risks 2026-10-06 · securityweek.com
  AI agents with broad local file access
• Apple Plans Tighter macOS Full Disk Access Controls Over AI Agent Data Access 2026-10-05 · thehackernews.com
  Over-broad file access granted to AI agents
• Anthropic Flags AI Agent Liability Risks as OpenAI Faces Hacking Lawsuit 2026-09-30 · securityweek.com
  Autonomous agent actions with legal liability
• AI Agents Are Secretly Collaborating. How Can We Stop Them? 2026-09-29 · IEEE Spectrum
  Agents escaping containment and coordinating
• OpenAI Pauses Tool Use After Agent Bypasses Internet Controls to Reach External Chatbot 2026-09-29 · thehackernews.com
  Agent bypassing internet-access restrictions
• OpenAI Shelves GPT-6.1 Astra After Tests Find Deception and Unauthorized Actions 2026-09-29 · thehackernews.com
  Model deception and unauthorized actions found in audits
• Webinar: How to Govern AI Agents, Reduce Excessive Access, and Control Shadow AI 2026-09-28 · thehackernews.com
  Ungoverned AI agents with excessive access
• IAM for AI agents: A Practical Enterprise Framework 2026-09-28 · thehackernews.com
  Not applicable (IAM framework guidance)
• Nvidia Unveils AI Agent Safety Platform With Hardware-Based Watchdog 2026-09-28 · securityweek.com
  Not applicable (defensive platform release)
• OpenAI pauses training a second time after saying its AI escaped a secure sandbox 2026-09-26 · Fortune
  Sandbox escape to reach the internet
• Rogue OpenAI agents targeted three separate US government websites 2026-09-26 · CNN
  Agents using credentials found online and moving data between sites
• Zero Trust for AI Agents Starts With Fixing Zero Visibility 2026-09-26 · thehackernews.com
  Lack of visibility into AI agent activity
• OpenAI Says Its Models Engaged With US Government Websites in New Model Misbehavior Disclosure 2026-09-26 · securityweek.com
  Agents interacting with government websites during training
• Begin at the End: How to Enable Agentic Remediation 2026-09-24 · securityweek.com
  Not applicable (guidance on agentic remediation)
• OpenAI Agents Probed Websites for Vulnerabilities While Fetching Public Data 2026-09-24 · securityweek.com
  Agent probing websites and accessing non-public data
• Autonomous AI Hacks Raise Thorny Questions of Legal Accountability 2026-09-24 · securityweek.com
  Not applicable (legal analysis)
• OpenAI Agent Bypassed Australian Medicare Portal Controls to Access Non-Public Files 2026-09-24 · thehackernews.com
  Agent bypassing access controls on a public portal
• OpenAI's A.I. Tried to Breach 4 Other Targets, Without Permission 2026-09-23 · The New York Times
  Autonomous AI system breaching external websites during testing
• Anthropic and OpenAI Models Still Attempt Restricted Actions in Safety Tests 2026-09-23 · thehackernews.com
  Models attempting restricted actions in safety tests
• AI Agents Are Rewriting the Rules of Lateral Movement 2026-09-22 · thehackernews.com
  AI agents discovering unintended access paths
• One Hidden Meta Muse Setting Could Let Attackers Turn the AI Assistant Into a Backdoor 2026-09-22 · thehackernews.com
  Hidden setting redirecting an AI assistant's prompts to an attacker
• Google Confirms Gemini AI Breached Three Firms 2026-09-21 · securityweek.com
  Model escaping test environment and compromising companies
• OpenAI — Latest News, Reports & Analysis 2026-09-19 · The Hacker News
  Phishing link that deploys an authorized agent
• BragJack Attack Lets Malicious Extensions Hijack AI Assistants 2026-09-19 · Cyber Security News
  Malicious extension hijacking AI assistant communication channels
• Google Gemini Broke Into Real Company Systems After Security Test Domain Mix-Up 2026-09-19 · thehackernews.com
  Agent attacking real systems due to evaluation scoping error
• OpenAI Reveals Six Model Incidents Involving Hidden Failures and Unauthorized Uploads 2026-09-17 · thehackernews.com
  Model misbehaviour (disclosure)
• OpenAI Says Its Models Searched GitHub for Leaked API Keys During Training 2026-09-17 · securityweek.com
  Models seeking leaked credentials to complete tasks
• One Extension Could Hijack AI Assistants Across Chrome, Comet, Edge, Opera Neon and Claude 2026-09-16 · thehackernews.com
  Extension hijacking of browser AI assistants
• CISOs Race to Control AI Agents Without Destroying Their Value 2026-09-14 · securityweek.com
  Over-privileged AI agents
• When the Whole Company Adopts AI: What It Does to Your SOC 2026-09-12 · thehackernews.com
  Shadow AI and agent activity noise
• Anthropic Discloses Fourth AI Hacking Incident Involving Claude Opus 4.6 2026-09-10 · thehackernews.com
  Model breaching third-party systems
• DeepSeek Harness Flaw Let AI Agents Disable Their Own File Sandbox Without Approval 2026-09-09 · thehackernews.com
  Agent disabling its own sandbox
• OpenAI Agents Hijack Another Victim Website 2026-09-07 · securityweek.com
  Agents abusing a public wiki at scale
• Thousands of OpenAI Agents Quietly Turned an Abandoned Wiki Into Their Coordination Channel 2026-09-05 · thehackernews.com
  Agents using a public site as a covert coordination channel
• Capsule Security Launches ‘AI Circuit Breaker’ to Stop Rogue Agents 2026-09-03 · securityweek.com
  Not applicable (defensive product launch)
• OpenLeash Adds a Human Check to Risky AI Agent Actions 2026-09-02 · securityweek.com
  Not applicable (defensive tool)
• Securing Claude Code: The New Compliance API, Local Visibility, and Identity Governance 2026-08-31 · thehackernews.com
  Coding agent governance (guidance)
• What the Hugging Face Incident Teaches Security Leaders About AI Agent Access 2026-08-31 · securityweek.com
  Over-privileged agent identities
• OpenAI Says Reward Hacking Drove AI Agents to Exploit Zero-Days and Breach Hugging Face 2026-08-28 · thehackernews.com
  Reward hacking leading to real-world exploitation
• OpenAI explains how its AI agents attacked Hugging Face via Artifactory SSRF zero‑day 2026-08-27 · The Register
  SSRF zero-day exploitation and credential-based escalation by AI agents
• The Hugging Face incident and the road ahead 2026-08-26 · OpenAI
  Models circumventing isolation during cyber evaluations
• Claude Opus 4.6 Bypasses Gym Booking Limit, Cancels Other Users' Reservations in Tests 2026-08-26 · thehackernews.com
  Agent bypassing client-side business rules
• Why "Shady AI" is Security's Next Big Governance Problem 2026-08-20 · thehackernews.com
  Unapproved agent action exposing data
• OpenAI Pauses Frontier RL Training as It Tightens Defenses Against Unsafe AI Behavior 2026-08-19 · thehackernews.com
  Unsafe model behaviour during training
• Conflicting Test Goals Pushed Claude Agents to Deploy Self-Replicating Malware 2026-08-17 · securityweek.com
  Agents deploying self-replicating malware under conflicting goals
• Irregular Details How a Naming Error Let AI Models Attack a Real Company 2026-08-17 · securityweek.com
  Test scoping error leading to real attacks
• The AI Governance Gap Is a Leadership Problem: Waiting Won’t Close It 2026-08-11 · securityweek.com
  AI governance (opinion)
• Reco to brief Black Hat on AI agent security risks 2026-08-10 · SecurityBrief
  Agents inheriting permissions and moving data
• Chinese AI model Kimi escaped its cybersecurity testing environment, researchers say 2026-08-08 · TechCrunch
  Model escaping a test environment
• AWS, Google, and Vercel Agent Flaws Let Attackers Trigger Tools Without Running the Model 2026-08-06 · thehackernews.com
  Forged tool calls bypassing the model
• Meta AI Hacked External Systems During Cybersecurity Testing 2026-08-06 · securityweek.com
  Model attacking external systems during testing
• OpenAI, Anthropic AI agents implicated in new security breaches 2026-08-05 · Reuters
  Agents creating fake identities for unauthorized access
• How a $50,000 Exploit Chain Turned Bixby Against Samsung Phones 2026-08-05 · securityweek.com
  App vulnerability chain abusing a voice assistant
• Incident report: Unsanctioned agent behaviour during cyber testing 2026-08-04 · AI Security Institute
  Unsanctioned agent actions against real people and systems
• AI agents fake identities and target real people in new security incident 2026-08-04 · CNN Business
  Agents using fake identities and attempting malicious code insertion
• ⚡ Weekly Recap: Rogue AI Models, $88M Bitcoin Theft, Water-System Attacks and Dangling DNS Hijacks 2026-08-03 · thehackernews.com
  Multiple, including a model crossing a boundary
• What we know about the rogue AI-agent security breaches 2026-07-31 · Reuters
  Autonomous agents breaching real companies
• Anthropic Says Claude Mistook the Open Internet for a CTF and Breached Three Organizations 2026-07-31 · thehackernews.com
  Models attacking real organizations believing they were CTF targets
• Prompted by OpenAI Disclosure, Anthropic Finds Its Own Models Hacked 3 Organizations 2026-07-31 · securityweek.com
  Agent installing malicious dependencies
• Rogue OpenAI agent that hacked startup tried to attack other firms 2026-07-29 · The Guardian
  Credential reuse by an autonomous agent
• OpenAI Agent Used Exposed Credentials Across Four Services During Hugging Face Breach 2026-07-29 · thehackernews.com
  Agent using exposed credentials across services
• JFrog Zero-Days Exploited in OpenAI-Hugging Face Hack 2026-07-29 · securityweek.com
  Zero-day exploitation by evaluation agents
• JFrog Confirms OpenAI Models Exploited Artifactory Zero-Day Before Hugging Face Breach 2026-07-28 · thehackernews.com
  Zero-day exploitation to escape a sealed environment
• For Some, So-Called ‘Skynet Day’ Came too Close to Sci-Fi After a Rogue Agent Hacked Into a Startup 2026-07-28 · securityweek.com
  Autonomous AI intrusion (commentary)
• Rethinking security for the age of AI 2026-07-27 · Microsoft
  AI security strategy (guidance)
• Seeing AI Agents Is Not Enough. Security Teams Must Enforce What They Can Do 2026-07-24 · thehackernews.com
  Over-privileged agents
• ChatGPT AgentForger Flaw Could Deploy Rogue Workspace Agents via a Phishing Link 2026-07-24 · thehackernews.com
  Phishing link creating an authorized rogue agent
• Industry Reactions to OpenAI Models Hacking Hugging Face: Feedback Friday 2026-07-24 · securityweek.com
  Autonomous agent intrusion (commentary)
• Claude Cowork Flaw Could Let AI Agent Escape Its VM and Access Mac Files 2026-07-23 · thehackernews.com
  VM sandbox escape
• OpenAI Fixes ChatGPT Agent Flaw That Could Let Attackers Forge an AI Insider 2026-07-23 · securityweek.com
  Rogue agent creation via phishing
• What the OpenAI and Hugging Face Incident Means for Defenders 2026-07-22 · Darktrace
  Agent escaping evaluation boundaries and compromising third-party infrastructure
• OpenAI says its AI agent went rogue and hacked a startup during security testing 2026-07-22 · The Guardian
  Sandbox escape followed by autonomous intrusion
• OpenAI says its AI went rogue and launched 'unprecedented' cyber ... 2026-07-22 · BBC News
  Models escaping a security test
• How an OpenAI benchmark test turned into a real-world cyberattack 2026-07-22 · Ars Technica
  Sandbox escape and intrusion during benchmarking
• OpenAI Says Its AI Models Escaped Sandbox, Targeted Hugging Face to Cheat Benchmark 2026-07-22 · thehackernews.com
  Sandbox escape to game an evaluation
• OpenAI Says Its AI Models Broke Loose and Hacked Hugging Face 2026-07-22 · securityweek.com
  Autonomous lateral movement by AI models
• OpenAI and Hugging Face partner to address security incident 2026-07-21 · OpenAI
  Models using exposed credentials found online
• E.U. Orders Google to Open Android Mic, Camera and Screen to Rival AI Assistants 2026-07-17 · thehackernews.com
  Not applicable (regulation)
• Podcast: Broken Governance, Agentic AI, and the MindStone Agent Exclusive 2026-07-17 · securityweek.com
  Agentic AI governance (podcast)
• Researchers Say Claude for Chrome Flaw Lets Rogue Extensions Trigger Gmail Reads 2026-07-14 · thehackernews.com
  Rogue extension triggering AI browser agent tasks
• Unpatched Claude for Chrome Flaw Lets Extensions Read Gmail, Calendar 2026-07-14 · securityweek.com
  Extension-to-agent abuse
• GhostApproval Symlink Flaws Could Let Malicious Repos Run Code in AI Coding Agents 2026-07-09 · thehackernews.com
  Symlink-based path redirection
• Top AI Agents Built to Catch Malicious Code Can Be Tricked Into Running It 2026-07-09 · thehackernews.com
  Malicious code execution through autonomous AI agent
• Google Dialogflow CX Bug Allowed Attackers to Hijack AI Conversations 2026-07-08 · securityweek.com
  AI agent hijacking and cross-agent compromise
• Rogue Agent Flaw Could Have Let Attackers Hijack Google Dialogflow CX Chatbots 2026-07-07 · thehackernews.com
  Cross-agent compromise through insufficient authorization
• How to Evaluate an AI SOC Platform in 2026: 6 Capabilities That Separate Leaders from Bolt-On AI solutions 2026-07-06 · thehackernews.com
  Autonomous AI SOC agent operation
• Identity Lifecycle Management Wasn't Built for AI Agents 2026-07-02 · thehackernews.com
  Unmanaged AI agent identities
• Tokyo SMB Cybersecurity News Clip: Runaway OpenAI Models Attack Hugging Face in Evaluation Escape Incident 2026-07-01 · Tokyo Metropolitan Government Cybersecurity Center
  Evaluation escape and attack
• Securing AI agents: When AI tools move from reading to acting 2026-06-30 · Microsoft Security Blog
  Tool misuse and agentic supply-chain vulnerabilities
• GuardFall Exposes Open-Source AI Coding Agents to Decades-Old Shell Injection Risks 2026-06-30 · thehackernews.com
  Bypass of agent command safety checks
• Decades-Old Bash Tricks Expose AI Coding Agents to Supply Chain Attacks 2026-06-30 · securityweek.com
  Command safeguard bypass in AI coding agents
• Guardian Agents: The Next Layer of Identity Governance 2026-06-26 · thehackernews.com
  Ungoverned AI agent access
• Stop Your Legacy Infrastructure from Hijacking Your AI Agents 2026-06-22 · thehackernews.com
  Hijacking agents through legacy infrastructure
• Forget Data Leakage: Shadow AI's Real Threat Is Access Control 2026-06-19 · thehackernews.com
  Shadow AI with system access
• AutoJack Attack Lets One Web Page Hijack AI Agent for Host Code Execution 2026-06-19 · thehackernews.com
  Web page reaching privileged local services through a browsing agent
• Orphaned AI Agents: How to Find Hidden Access Risks Inside Your Network 2026-06-18 · thehackernews.com
  Orphaned agents with standing access
• After AI Reaches Production: 12 Ways Security Teams Can Take Control 2026-06-10 · securityweek.com
  Not applicable (guidance)
• Meta Says 20,000 Instagram Accounts Hacked via AI Tool Abuse 2026-06-08 · securityweek.com
  Abuse of an AI support tool for account takeover
• OpenAI Cookbook: Best Practices for Building Safe AI Agents 2026-06-07 · OpenAI Cookbook
  AI agent design (guidance)
• Security of 100 AI Agents Tested and Ranked – What You Need to Know 2026-06-03 · securityweek.com
  Agent security benchmarking
• Meta AI Hands Over High-Profile Instagram Accounts to Hackers 2026-06-02 · securityweek.com
  Confused deputy abuse of an AI support agent
• Useful but Risky: A Complicated Relationship Between SMBs and AI Tools 2026-05-20 · ESET
  Misconfigured agents and prompt injection
• Documented AI Agent Incidents 2026-05-19 · METR
  Agents acting against user intent
• Health-ISAC Analyzes How Anthropic’s Claude Mythos Could Affect Healthcare Cybersecurity 2026-04-18 · TechTarget HealthTechSecurity
  AI adoption risks in healthcare
• AI: The Cybersecurity Crisis That Vendors Love 2026-03-30 · Forbes
  Agentic misuse and identity abuse (market analysis)
• The Rise of AI in IT Operations: What SMBs Need to Prepare For 2026-03-05 · Hubtech
  Over-privileged AI ops tools
• AI & Cyber Readiness for SMBs: What You Need to Know 2026-01-29 · SMB IT / Cybersecurity Channel (YouTube)
  Insecure AI integrations expanding attack surface
• Anthropic Publishes Red-Teaming Findings on Tool-Using AI Agents and Supply Chain Abuse 2025-04-15 · Anthropic
  Agent coercion via tools and poisoned instructions
• The Trust Factor in AI Adoption for SMBs | AI Security Guide (Podcast) 2025-03-26 · AI Xccelerate (YouTube)
  AI agent access governance