中/EN
Flash

Flash·Safety & Risk·2026-10-18 20:06

OpenCode Hit by "Distillation Attack": API Keys of About 1.4% of Users Compromised

This time the attackers were not after passwords or credit card numbers. Open-source AI coding platform OpenCode admitted publicly on October 17 that attackers exploited a vulnerability in one of its legacy systems to gain unauthorized access to third-party model provider API keys stored in user workspaces.

HIMIAO Horizon·2026-10-18 20:06·1 min read

Listen Save Share card Copy link
Bright illustration of a laptop code screen with an API-key security warning
Illustration of the OpenCode distillation attack (AI-generated)

What happened

OpenCode says the incident affected about 1.4% of its Console (hosted workbench) users who had linked external AI provider accounts. The keys were connected to models from Anthropic, OpenAI and Google. With the keys in hand, the attackers ran repeated queries against top-tier models; affected users reported their account balances draining quickly, with some bills climbing into the hundreds of dollars.

OpenCode characterized the incident as a large-scale "distillation attack": the attackers used the stolen keys to harvest high-quality outputs from top-tier models to train their own models, with the bills landing on the victims. Distillation describes a technique where one AI model learns by studying the answers of another.

OpenCode says it is working with Anthropic, OpenAI and Google on remediation, and advises all users who linked external provider accounts to revoke and rotate their API keys and review usage on each provider's dashboard. How the unauthorized charges will be handled has not been specified by OpenCode or the three providers.

Key facts

  1. Scope: About 1.4% of Console users — those who linked external model provider accounts
  2. Entry point: A vulnerability in a legacy system being retired, not the main service
  3. Keys in scope: API keys for Anthropic, OpenAI and Google models were all in scope
  4. Method: "Distillation attack": harvesting top-tier model outputs with stolen keys to train a competing model
  5. User impact: Some users' bills hit hundreds of dollars overnight; who covers the charges is unclear

Why it matters

What was stolen was not data but billable frontier-model access — a single key can yield training-data value. The convenience of storing keys for multiple providers in one workspace concentrates the risk along with it.
Useful Tap if this story helped you

SourcesCrypto Briefing (Oct 17, 2026, https://cryptobriefing.com/opencode-distillation-attack-api-key-breach/).This is a compilation of public information and does not constitute investment advice.

From HIMIAO Horizon · himiao.ai

OPENCODEAPI KEYSECURITY

Comments

  1. Loading comments…

Related

  1. Oct 13MI5 Announces AI Threat Assessment Centre as Chief Warns AI Is Supercharging National Security Threats
  2. Oct 13Anthropic Says Anti-Distillation Safeguards Cut Extraction by 90% or More
  3. Oct 6Anthropic Expands Cyber Verification Program: After 129,000 Bugs Found, Its Strongest Models Open to Vetted Defenders — With Fewer Safeguards
  4. Oct 2OpenAI Disrupts "Distillation" Extraction Campaign: 16,000 Requests Point at Moonshot-Linked Actors
  5. Oct 17OpenAI Wins Pentagon Classified-Systems Deal, Agreeing to "All Lawful Purposes" With Its Own Technical Safeguards

Today

Oct 19 Monday

Latest news

All →
  1. Oct 18Anthropic to Commit 5% of Shares to New Public Benefit Fund
  2. Oct 19Stably AI's Orca Tops GitHub Daily Trending: One Desktop App to Orchestrate 30+ Coding Agents
  3. Oct 19Jailbroken PS5 Becomes a Local AI Server: Qwen3.8-27B Hits ~20 tok/s, Small Models Top 100
  4. Oct 19Anthropic Internal Study: Junior Share of Technical Staff Falls to 14%, Meeting Time Up 20%
  5. Oct 18Claude Cracks 25-Year Cryptography Mystery in Half an Hour: NIST Curve Seeds Came From a 1997 Clock
  6. Oct 18Websites Are Closing Their Doors to AI Agents: Yelp, Delta, JPMorgan Draw Lines
Ask the cat