Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Symantec Threat Hunter Team hero artwork for its unauthorized AI relay report, a circuit-board style illustration with a glowing core.
Image: Symantec / Broadcom (fair use).

Detection

Symantec Says an Unauthorized Relay Installs Real Codex and Claude Code CLIs, Then Sets sandbox_mode to danger-full-access and Routes Every Prompt Through Its Own Servers

Threat Hunter found the grey-market service on systems at a dozen organizations since early 2026, mostly in mainland China. A fileless setup script fetched from lookalike domains rewrites CLI config, replaces API credentials and disables telemetry; Symantec published hashes and domains and said it has not yet seen the operator abuse the access.

Ava Okello, DetectionLondon6 min read

LONDON - Symantec's Threat Hunter Team said Thursday that an unauthorized relay service has been installing the real OpenAI Codex and Anthropic Claude Code command-line tools on enterprise systems, then rewriting their configuration so every prompt and credential is routed through third-party servers with the tools' execution sandboxes switched off.

The service has been active since at least early 2026 and has turned up on systems at a dozen organizations, most of them in mainland China, along with multinational subsidiaries in South Korea, Taiwan and Malaysia. Affected sectors include industrial automation, electronics manufacturing, semiconductors, pharmaceuticals, shipping, tire manufacturing, petrochemicals and data center services. Symantec said there is no evidence so far that the operator has abused the access, but the configuration it installs would let it read every prompt, file and secret the agent sees and run commands with the user's full privileges.

Real tools, rewritten config

The service does not ship a trojanized client. Users run a legitimate npm install of @openai/codex, codex-cli or @anthropic-ai/claude-code shortly before the override, within minutes to roughly half an hour. A fileless PowerShell or bash setup script, fetched with a one-line iwr -useb | iex or curl | bash from a lookalike domain such as vibe.aiok[.]me, checks for Node.js and the target CLI, installs them if needed, and only then applies its own configuration. The script prints progress messages in Chinese and tells the user to re-run it elevated if it lacks administrator rights, which Symantec said indicates users are copying a ready-made command from the relay or a reseller.

Key findings section of Symantec's Threat Hunter Team report on unauthorized AI relays.
Image: Symantec / Broadcom (fair use).The report's key findings block.

The script rewrites ~/.codex/config.toml for Codex or ~/.claude/settings.json for Claude Code. It replaces the base_url or ANTHROPIC_BASE_URL with a domain the relay controls, swaps in a relay-issued API key, and sets both as user-scope environment variables so the change survives a reboot. For Codex it sets sandbox_mode to "danger-full-access," enables network access for the write sandbox, and turns off telemetry and error reporting. On at least one host, Symantec said, the script ran claude /logout, deleted the ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY environment variables, and then ran claude /login before installing its own credentials.

Configuration override section of the Symantec report, including sandbox_mode set to danger-full-access.
Image: Symantec / Broadcom (fair use).The config and credential changes the setup script makes.

Several hosts also showed the script querying the Windows locale setting immediately after installation, which Symantec said is consistent with a service confirming it is running in the region it serves. Claude Code keys issued by the relay carried the sk-ant-oat01- prefix Anthropic uses for subscription-based OAuth tokens rather than pay-per-use API keys, which is consistent with, though not proof of, a relay pooling flat-rate subscriptions.

What SOCs should watch

Symantec was careful about attribution of intent. Bypassing regional restrictions on Western AI vendors is the simplest explanation, it said, but that does not require disabling a CLI's sandbox or planting persistent credentials. Fraud involving stolen credentials or payment cards, and resale of the prompts and files that pass through the relay, are also plausible and cannot be ruled out. "This is not a vulnerability in Codex or Claude Code," the report stressed. "The tools are behaving as designed once their configuration has been changed."

Protection and indicators of compromise section of the Symantec report.
Image: Symantec / Broadcom (fair use).File hashes, lookalike domains and the endpoint checks Symantec recommends.

Organizations should watch for changes to base_url or ANTHROPIC_BASE_URL in a CLI's configuration file, sandbox_mode set to "danger-full-access" in ~/.codex/config.toml, OPENAI_BASE_URL, CODEX_API_KEY or ANTHROPIC_AUTH_TOKEN set as user-scope environment variables outside an approved process, a PowerShell iwr | iex or bash curl | bash invocation run shortly after or instead of the legitimate npm installs, and traffic from AI CLI tools to destinations outside an allow-list of approved AI endpoints. Symantec published SHA-256 hashes for setup-codex.ps1 and setup-claudecode.ps1 and a list of lookalike domains that have included vibe.aiok[.]me, aiok[.]me, sorryios[.]io, nuoda[.]vip, rawchat[.]cn, sharedchat[.]cc, mirbuds[.]com and claudechn[.]com. Because the credentials are written as user-scope environment variables, any other tool on the same machine that reads those variables can pick them up.

Sources:


Ava Okello covers detection engineering, EDR telemetry, and SOC hunting for SOCtember from London.

Related stories

Detection desk