AI Coding Tools Under Fire: Grok Build Uploads Entire Git Histories, Claude Code Suspected of Silent Transfers
Security researcher cereblab exposed how Grok Build and Claude Code engage in unauthorized data collection that goes far beyond user expectations, turning routine code assistance into large-scale repository exfiltration.
In a controlled test with a 12 GB local Git repository, Grok Build 0.2.93 created two independent HTTPS channels. The primary channel handled task context at roughly 192 KB, while a background storage channel silently repacked and uploaded the entire .git directory in 75 MB chunks, consuming 5.10 GiB of traffic to a Google Cloud Storage bucket named grok-code-session-traces. This produced a 27800-fold difference between expected and actual outbound volume.
The upload logic activated regardless of explicit user instructions such as "do not read" placed on marker files, proving that the behavior is a hardcoded data-collection routine rather than an artifact of the model's understanding. Privacy controls proved ineffective: the client-side improve_model_enabled flag is completely decoupled from the server-side trace_upload_enabled flag, allowing uploads to continue even after users disable model-improvement features.
Although xAI later disabled the uploads via remote configuration on 13 July, the full repository-upload code remained present in the 0.2.99 binary, merely gated by a server flag. The exposed Git history includes deleted .env files, credentials, connection strings, and commit records that reveal vulnerability-fix timelines, enabling attackers to infer unpatched issues in similar codebases.
Parallel findings emerged for Claude Code, whose client maintains multiple WebSocket connections that periodically transmit file paths, dependency trees, and code fragments even without an active coding task. No local logs, switches, or data-flow documentation are provided, and the closed-source binary prevents independent verification of additional hidden routines.
A side-by-side traffic audit of major tools produced clear results:
- Grok Build: full .git directory, Git history packaged, privacy switch ineffective, confirmed silent upload.
- Claude Code: suspected file-level and metadata exfiltration, details undisclosed, silent-transfer behavior confirmed.
- Codex: only current context fragments, no Git packaging, privacy controls effective, no anomalous reports.
- Gemini: no active uploads detected.
Analysts attribute the pattern to fundamental conflicts between commercial model-training demands and user data sovereignty, excessive server-side control over local clients, and the absence of mandatory third-party audits for closed-source AI binaries.
Recommended defenses include physical network blocking of known endpoints, running tools inside Docker containers that mount only read-only directories without .git, pre-processing repositories with git filter-repo, deploying eBPF monitoring with alerts for sessions exceeding 10 MB, and favoring auditable open-source options such as Continue.dev or locally hosted Ollama models.
Related articles
Israeli Firm Reveals First Known AI-Led Breach of Taiwanese Government Systems
An Israeli cybersecurity company named Dream discovered an open 160 MB archive containing 1,395 files that documented a fully autonomous AI operation against Asian government infrastructure later identified as Taiwan. Between July 1 and July 4 2026 the system ran 12 sequential waves using up to eight sub-agents simultaneously, each handling reconnaissance, exploitation, lateral movement and persistence without further human input after initial setup. The agents mapped 21 interconnected government systems, exploited unauthenticated debug endpoints and single-sign-on weaknesses, and ultimately compromised 85 employee accounts while exfiltrating more than 2,500 personnel records. The framework relied exclusively on two publicly available open-source AI assistants, Hermes and OpenClaw, and bypassed model safety filters by framing the task as an authorized penetration test. The same agents later expanded into government IT suppliers, the national email system, seven energy companies and the nuclear safety agency while performing internal validation that rejected seven false-positive findings. No zero-day exploits were used; all successful access paths involved exposed endpoints, disabled signature checks and missing authentication controls.
GitHub Copilot Traffic Analysis via MITM Proxy Exposes Prompt Context Handling and Local SQLite Session Storage
A detailed reverse-engineering study placed GitHub Copilot behind an mitmproxy instance to inspect all network requests made by Visual Studio Code. The analysis revealed that Copilot performs OAuth token exchange, model availability checks, and intent classification before any user input occurs. Prompts sent to the model include context from recently edited files, even when inline suggestions are disabled for sensitive extensions such as .env. Copilot maintains a local SQLite database named session-store.db that records every user prompt, LLM response, repository, and branch worked on. The extension also exposes a session_store_sql tool allowing the model to run read-only SQL queries against this history using the Copilot Chronicle skill. These findings highlight how AI coding assistants manage context, authentication, and persistent local state.
Anthropic Rolls Out Invisible Statistical Watermarks for Claude Models to Comply with EU AI Act
Anthropic has embedded invisible statistical watermarks into all outputs from its Claude models starting August 2, 2026, to meet Article 50 of the EU AI Act. The two-layer system applies a token-level bias using a secret key for text and C2PA metadata for images and files. Open-source projects appeared within 24 hours promising to strip the marks, yet none have demonstrated verifiable success against the statistical layer because Anthropic has not released a public detector. The technique, first described by Kirchenbauer et al. in 2023 and deployed by Google as SynthID, works by subtly biasing token selection toward “green” lists during generation. Editing, translation, or full paraphrasing rapidly degrades detectability, while short or rigidly formatted text such as code offers little room for the signal. The move affects every Claude deployment worldwide, not only EU users, to avoid maintaining dual model versions.
Guardrails Filter Tackles Complex LLM Streaming and Tool Call Challenges to Protect Sensitive Data
Developers at Cloud.ru built Guardrails Filter to mask personal data such as phone numbers, emails, passport details and names before they reach large language models. The system replaces detected values with consistent placeholders like <PHONE_1> and maintains a mapping table so original data can be restored after the model responds. Simple replacement proved insufficient because identical values must receive the same placeholder across an entire conversation history, and the model receives the full message array on every request. Streaming responses using SSE create additional difficulties since placeholders can be split across multiple chunks, requiring buffering of 10-15 characters and state tracking for reasoning, content and tool_calls. The team also had to handle JSON-inside-JSON arguments for tool calls, different field names across providers, and edge cases such as escaped newlines matching email patterns. Separate implementations were written for OpenAI Chat Completions and Anthropic Messages APIs, resulting in roughly 1,500 lines of streaming code and more than 4,000 lines of tests to ensure agent pipelines remain intact.