OpenCode Got Hacked. Anthropic and OpenAI Refused to Help.

Attackers broke into OpenCode, one of the most popular AI coding tools, and people woke up to AI charges they never made. Then, in the middle of a 48-hour fight, the frontier AI OpenCode was using to defend itself started refusing to help. It's the second time this year, after Hugging Face. Here's what happened, what we don't know yet, why the fix is a permission slip, and what you should do with your AI keys this week. Free AI assessment for your business (about 10 minutes): https://rmdw.ai/?utm_source=youtube Become an RMDW Insider ($4.99/mo): every essay 24 hours early, a vote on the next topic, your name in the credits: https://www.youtube.com/channel/UCvZNH8z8q38CJ3Cqe1PMLCw/join Follow me on X: https://x.com/richardbechols CHAPTERS 0:00 OpenCode got hacked 0:58 Chapter 1: The break-in 3:03 Chapter 2: What distillation means 4:41 Chapter 3: The AI said no 8:34 Chapter 4: We've seen this before 12:33 Chapter 5: The fix is a permission slip 17:09 Chapter 6: Why this is your problem 18:44 Chapter 7: What to do this week DO THIS WEEK 1. Revoke old AI keys at OpenAI, Anthropic or Google directly. Deleting a key inside another app does not kill it. 2. Set a spending limit on every AI account. 3. Use one key per tool, so if one leaks you know which. SOURCES OpenCode breach statement (1.4% of Console users, "large scale distillation attack"): @opencode on X (Oct 17, 2026) Dax Raad posts (48 hours of attacks, agents 24/7, models "started refusing", Daybreak Blue, "i don't think anyone is refusing to help us"): @thdxr on X (Oct 18, 2026); OfficeChai (Oct 18, 2026) User charges ($281 and $168): @colinparsonscom on X; Remy.Blog (Oct 18, 2026); Notebookcheck (Oct 17, 2026); Crypto Briefing; The420.in Anthropic revoked affected Claude keys: Lydia Hallie on X (Oct 18, 2026) Hugging Face attack and refusal, GLM 5.2 on its own servers: Fortune (Jul 20, 2026); OpenAI models escaped testing and hacked Hugging Face (ExploitGym): Fortune (Jul 21, 2026) Daybreak Blue and Red: OpenAI; AWS Machine Learning Blog (Aug 11, 2026); Artificial Analysis Cyber Index (Oct 8, 2026) Dean W. Ball and David Sacks posts on X (Oct 18, 2026); Sacks' role: The Hill via Yahoo News (Mar 27, 2026) Andrew Curran on X (Oct 18, 2026); Digg (Oct 18, 2026) OpenCode and Hugging Face have not publicly named which models refused; the opinions in this video are Richard's own. RMDW = Real-world Models, Data and Workflows. AI is about results, not prompts. CREDITS News and creator footage (CNBC, CBS News, CBS Mornings, NBC News, CNN, Bloomberg, The Wall Street Journal, Face the Nation, Fox 5 NY, CBS4 Indy, CBS New York, India Today Global, Reuters, OpenAI, Anthropic, The Pragmatic Engineer, DevOps Toolbox, The Intelligence Desk, Code Bear, Superbash) used briefly, muted, for commentary and news under fair use. Logos are trademarks of their owners, shown for commentary. KiyomiKode Desktop footage: RMDW. #AI #OpenCode #Cybersecurity #Anthropic #OpenAI