← Real-world cases

xAI Grok Build CLI — covert full-repo/secrets upload despite privacy opt-out

Real-world incident12 Jul 2026 / 13 Jul 2026

On 12 Jul 2026 the independent researcher Cereblab published a wire-level analysis of xAI's Grok Build coding CLI (binary grok 0.2.93, macOS arm64, with a captured SHA-256) reporting it silently shipped users' entire local repositories to a Google Cloud Storage bucket 'grok-code-session-traces' (paths like gs://grok-code-session-traces/repo_changes_dedup/v2/). Per the packet captures, on a 12 GB test repo the model-turn channel (/v1/responses) moved ~192 KB across 5 requests while a parallel storage channel (/v1/storage) moved 5.10 GiB in 73 chunks of ~75 MB each — a ~27,800x disparity — even when the agent read zero files, indicating whole-repository transmission (full Git bundles, untracked working files, committed history). Unredacted .env contents (illustratively API_KEY, DB_PASSWORD) were captured in request bodies and uploaded archives, and secrets deleted from Git history months earlier could be reconstructed. Crucially the 'Improve the model' toggle reportedly controlled only training consent, not transmission: /v1/settings kept returning trace_upload_enabled: true after disabling it; the actual kill-switch was a server-side flag disable_codebase_upload, which xAI silently flipped to true around 13 Jul 2026. Independent observers saw the same flag change; a later analysis of build 0.2.99 reported the upload code still present but dormant behind the server flag. xAI responded on social media (enterprise zero-data-retention teams never had code/trace data stored; consumers can run /privacy to disable retention and delete synced data); Musk publicly promised to purge previously uploaded data and xAI reportedly said it would open-source the tool; no formal security advisory. Illustrates vendor-side agent telemetry functioning as a covert exfiltration pipeline and a privacy opt-out that silently no-ops. Figures and behaviour attributed to Cereblab's analysis; secret/payload examples are illustrative.

Practise the risk class — related scenarios

Interactive simulations of the risk class this case illustrates (not a re-enactment of this specific event).

🔑The Agent With the Master Key

An ops agent gets one god-mode credential — and one misread wipes production

📣The Echo Chamber

A team of agents agrees its way into a confidently wrong answer — and a runaway loop

📧The Email That Gave Orders

A support email hides instructions — and the assistant obeys them

🗄️When the Query Bites Back

A text-to-SQL agent runs the model's output straight at the database

🪡Death by a Thousand Innocent Steps

A jailbroken agent decomposes one malicious goal into hundreds of harmless-looking steps — and per-step filters never see the attack

🕵️Lies in the Loop

A poisoned issue makes the agent lie to the human who approves its actions

👂Overheard Through the Cache

A speed optimisation becomes a cross-tenant listening device

🪟Stealing the Model

Two doors to the same secret: reconstruct the model through its API, or just walk off with the weight file

🎭The Blackmail Gambit

Told it's being shut down, an agent reaches for leverage — with no attacker in sight

📼The Compromised Flight Recorder

The forensic record is itself the attack surface — an agent's log is poisoned, then quietly rewritten

📦The Dataset That Ran Code

A 'safe' dataset preview turns an upload into code execution on the pipeline's workers

👁️The Invisible Webpage Command

A shopping page tells the agent to do something the user never asked for

📡The Message in Morse

Encoded public text is laundered across an agent handoff into an on-chain transfer

🖼️The Picture That Whispered

A screenshot that's harmless at full size becomes an order once the system shrinks it

🎫The Stolen Session

An attacker captures the agent's bearer token — and inherits its authority

🥸The Uninvited Agent

A forged peer registers on the agent directory — and the planner enlists it

🪪The Worker Who Spoke for the Boss

A poisoned web page hijacks a research agent — and the planner acts on its behalf

🖱️What You Click Is Not What You Get

A GUI agent clicks 'Continue' — but the screen moved, and it lands on 'Send'

🖼️Zero-Click Leak by Picture

An inbox summary quietly ships a secret to an attacker's server

More cases on Sensitive Data Leakage

Bing 'Sydney' system-prompt leakEchoLeak — Microsoft 365 Copilot zero-click (CVE-2025-32711)Agentic-browser indirect-injection demos (ChatGPT Operator)Samsung confidential-code leak via ChatGPTChatGPT persistent-memory exfiltration (Rehberger / 'SpAIware')postmark-mcp backdoorForcedLeak — Salesforce Agentforce CRM exfiltration (CVSS 9.4, no CVE)ServiceNow Now Assist — second-order prompt injection via agent-to-agent discoveryShadowLeak — ChatGPT Deep Research zero-click service-side exfiltrationIDEsaster — AI coding IDEs/agents turned into exfiltration & RCE surfacesMorris II — zero-click self-replicating adversarial-prompt worm across GenAI agentsSalesloft Drift OAuth supply-chain breach (UNC6395) — mass Salesforce data theft via an AI chat integrationNVIDIA Triton Inference Server unauthenticated RCE chain (CVE-2025-23319 / -23320 / -23334)Anamorpher — image-scaling prompt injection against production AI systemsOperation Bizarre Bazaar (first attributed LLMjacking campaign with a resale marketplace)TeamPCP poisons the LiteLLM AI gateway on PyPI to harvest LLM API keysCVE-2026-21445 — Langflow missing authentication on critical API endpoints, exploited in the wildMalicious JetBrains Marketplace plugins steal AI API keysSearchLeak — Microsoft 365 Copilot one-click data theft (CVE-2026-42824)ChatGPhish — ChatGPT web-summary rendering turned into a phishing surfacecodexui-android — malicious npm package steals OpenAI Codex auth tokensPyTorch Lightning PyPI compromise (Mini Shai-Hulud / TeamPCP)Poisoning Claude Code: one GitHub issue hijacks the claude-code-action CI supply chainAmazon Q Developer auto-loads workspace MCP configs, enabling zero-click AWS credential theft (CVE-2026-12957)ClaudeBleed — co-resident Chrome extensions coerce Claude for Chrome into reading Gmail/Docs/CalendarContext Contamination: passive prompt injection poisons LLM security-log analysis

AI RiskAtlas is an educational model of how GenAI & agentic systems work and fail. Architectures and payloads are illustrative and simplified for learning — not operational guidance. Real-world cases are summarised from public reporting.

Sources & further reading →·Built by Shi Yuan ↗