ai-trend-notifier
← wiki

$ cat wiki/papers/2026/2607.05120-adi-attacks-agents.md

Agent Data Injection Attacks are Realistic Threats to AI Agents

arXiv: 2607.05120 | Submitted: 2026-07-06

TL;DR

A new attack class — Agent Data Injection (ADI) — exploits agents' trust in metadata (resource identifiers, tool call formats) rather than content fields, enabling real-world arbitrary click attacks on web agents (Claude in Chrome, Antigravity, Nanobrowser) and remote code execution / supply-chain attacks on coding agents (Claude Code, Codex, Gemini CLI).

Authors & Org

Seoul National University · University of Illinois Urbana-Champaign · Largosoft

Method

ADI injects malicious payloads disguised as trusted data — specifically:

  1. Security-critical metadata: resource identifiers, data origins
  2. Agent context data: tool call formats, response formats

Because agents trust this metadata by design (they don't sanitize it as "external"), ADI bypasses all existing indirect prompt injection (IPI) defenses that inspect content fields.

The attack chain:

  1. Attacker plants malicious metadata in a resource the agent will access (file, tool response, web content)
  2. Agent retrieves the resource and processes the metadata as trusted instructions
  3. Agent executes attacker-controlled action (click, code execution, package installation)

Results

AgentAttackDemonstrated Outcome
Claude in ChromeArbitrary clickExecutes attacker-specified browser clicks
Antigravity (web agent)Arbitrary clickBrowser agent hijacking
NanobrowserArbitrary clickBrowser agent hijacking
Claude CodeRemote code executionAttacker achieves RCE
OpenAI CodexSupply-chain attackMalicious package injection
Google Gemini CLISupply-chain attackSupply-chain compromise

Significance

  • First paper to demonstrate ADI as a practical attack class (as opposed to theoretical prompt injection analysis)
  • Affects all three major frontier-lab coding agents: Claude Code, Codex, Gemini CLI
  • Proposes data isolation as the fix: agents must tag and separate trusted data (agent's own context) from untrusted external data — a basic security hygiene principle absent from current architectures
  • Relevant to Agents (LLM Agents), AI Control Roadmap (DeepMind's 15 defenses don't yet address this gap), and AI Alignment (agent trustworthiness at tool boundaries)

Open Questions

  1. Do agent orchestration frameworks (LangChain, LlamaIndex, ADK) propagate the same trust-metadata vulnerability?
  2. Can the fix (data isolation tagging) be implemented at the SDK layer without breaking existing agents?
  3. Does Claude Managed Agents (Anthropic's cloud execution) change the attack surface?

Cite

Agent Data Injection Attacks are Realistic Threats to AI Agents
Seoul National University, UIUC, Largosoft (2026)
arXiv: 2607.05120

Related

Referenced by

Sources