AI Governance Institute
← News
Research2026-11-02

Prompt Injection Techniques Leaked PII and Ran Unauthorized Code in AI Agents

What happened

Adversa AI published its AI agent security incidents and vulnerabilities, October 2026 roundup. It documented prompt injection techniques that caused vision-language models (AI systems that process both images and text) to emit unauthorized tool calls and expose personal data. Separate demonstrations showed agents executing attacker-supplied code, extracting stored credentials, and accessing data belonging to other customers on shared platforms. Three underlying failures were consistent. Agents had been granted more access than their tasks required. Commands were not adequately checked before execution. Isolation between agent sessions was incomplete. The report reinforces patterns identified in prior research, including ACM Research Maps Five Privacy Leakage Pathways in Multi-Agent AI and MITRE ATT&CK v20 Names Prompt Injection and AI Agent Abuse as Formal Threat Categories. These attack classes are now documented, named, and actively exploited.

Why it matters

  • ·Cross-tenant data exposure, where one customer's agent accesses another's data, can trigger breach notification obligations under the General Data Protection Regulation (GDPR) and equivalent state laws. A single misconfigured agent privilege can create a notifiable incident at scale.
  • ·Regulators including the UK Information Commissioner's Office and US Federal Trade Commission have signaled that inadequate agent controls constitute foreseeable harm. Organizations that have not documented privilege boundaries and validated tool-call behavior face enforcement exposure when incidents occur.
  • ·The Five Eyes Guidance on the Careful Adoption of Agentic AI Services explicitly calls for least-privilege access and runtime behavioral controls. Enterprises without these in place now have named guidance they are failing to follow, which becomes relevant in any post-incident regulatory review.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Map every deployed AI agent to the systems, credentials, and data it can currently reach, and remove access that is not required for its defined task.
  • ☐Confirm with your engineering team that agent tool calls (the commands an agent sends to external systems or internal APIs) are validated before execution, not only after the agent receives a response.
  • ☐Ask whether your AI platforms isolate each agent session so that one agent cannot read data written or retrieved by another agent on the same platform.
  • ☐Review vendor contracts for AI agent platforms to verify that the vendor is required to notify you of incidents involving unauthorized data access across customer boundaries.
  • ☐Add prompt injection to your next AI red-teaming exercise and require evidence that the test covered image-processing agents, not only text-only tools.

What to watch next

Regulators are moving from guidance to enforcement on agentic controls. The EU AI Act (Regulation (EU) 2024/1689) and its enforcement framework are now fully in force for general-purpose AI. The European Commission has begun inspections targeting AI systems in hiring, credit, and healthcare contexts. US state laws including the Texas Responsible AI Governance Act and Colorado's forthcoming Colorado SB 26-189: Automated Decision-Making Technology Act will add further obligations for enterprises running agents that affect individuals. Compliance teams should monitor whether Adversa AI's documented techniques appear in future FTC or data protection authority enforcement actions. Named, published attack methods make it harder to argue that risks were unforeseeable.

Related Coverage

Standards2026-11-16

Microsoft Defense Guide Makes Layered Prompt Injection Controls an Enterprise Baseline

Microsoft has published updated guidance on defending against indirect prompt injection, a category of attack where malicious instructions hidden in external content redirect AI systems to perform unintended actions. The guidance, titled 'Defend against indirect prompt injection attacks,' recommends a layered set of controls including prompt shields, content isolation, behavioral monitoring, and human approval gates. It applies to any organization using AI systems that process third-party content such as emails, documents, websites, or plugins.

Research2026-11-14

One Rogue Employee-Built AI Agent Exposed Five Identity and Control Gaps

Token Security published a case study tracing how a single AI agent created five governance failures. The study draws on a Cloud Security Alliance survey of 418 IT and security professionals. The failures span agent identity, access provisioning, prompt injection, overprivileged static credentials, and behavior-based detection. The findings apply to any enterprise where employees can build agents using no-code or AI-assisted tools.

Corporate Policy2026-11-22

Treasury Nominee Leaves Agentic AI Bank Liability Unresolved at Senate Hearing

A Treasury Department nominee declined to define who bears responsibility when a third-party AI agent takes unauthorized actions in a consumer bank account. The nomination hearing before the Senate Banking Committee on November 20, 2026 produced no regulatory commitment beyond a pledge to coordinate with existing bank regulators. The Bank Policy Institute separately stated that agentic AI risk management is not sufficiently developed to support account access, naming prompt injection as a concrete unresolved threat.