Short answer: On November 9, 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. They are the same underlying model with different safeguards. Fable 5.2 is available to everyone on paid Claude plans, the Claude Platform and the big clouds; Mythos 5.2 is only for vetted security and life-sciences organizations. Fable 5.2 costs $8 / $40 per million input / output tokens, runs up to 30% faster than 5.1, and scored 49.1% on FrontierCode 2.0. In the same week, security firm Apiiro reported that Claude Code’s “Yes, and don’t ask again” option had put passwords and API keys into public GitHub repos through .claude/settings.local.json. If that file is tracked in your repo, untrack it and rotate the secret.
Claude Fable 5.2 vs Claude Mythos 5.2: one model, two doors
Anthropic’s launch post is clear on this point: Fable and Mythos are the same underlying model. The difference is the safeguards around it.
| Claude Fable 5.2 | Claude Mythos 5.2 | |
|---|---|---|
| Safeguards | Strong cybersecurity and biology safeguards | Some of those safeguards relaxed |
| Who can use it | Everyone on paid Claude plans | Vetted organizations only |
| Where | Claude apps, Claude Code, Claude Platform (claude-fable-5-2), AWS, Google Cloud, Microsoft Azure | Cyber Verification Program, Life Sciences Verification Program, or Claude Security on Enterprise |
| Best for | The biggest, most open-ended technical jobs | Vulnerability research and biomedical work |
When Fable’s safety classifiers block a high-risk cybersecurity or biology request, you get an answer from an older Opus model instead. The good news for everyday developers: in early testing Fable 5.2 blocked about 55% fewer requests than Fable 5.1 (about 60% fewer in cybersecurity and 40% fewer in biology).
The numbers that matter
Price
Fable 5.2 and Mythos 5.2 cost $8 per million input tokens and $40 per million output tokens, down from $10 / $50 for the 5.1 models. An optional fast mode speeds up generation by up to 2.5x for $16 / $80.
Speed
Anthropic says outputs come up to 30% faster than Claude Fable 5.1 and Mythos 5.1.
Coding score
On FrontierCode 2.0, a test of whether a model’s code changes would actually be merged by open-source maintainers:
| Model | FrontierCode 2.0 |
|---|---|
| Claude Fable 5.2 (Xhigh effort) | 49.1% |
| Claude Opus 5.5 | 45.5% |
| GPT-6 Astra | 42.8% |
| Gemini 4 Carbon | 41.1% |
| Claude Fable 5.1 | 32.4% |
Anthropic says 49.1% is the highest score of any released model it has tested. The more practical number: at High effort, Fable 5.2 scores about the same as Opus 5.5’s best (45.3% vs 45.5%) for roughly half the cost per task.
Which model should you use?
Anthropic still recommends Opus 5.5 as the daily driver for most coding and knowledge work, then moving up to Fable for the most ambitious jobs. My take as a Laravel and Next.js developer is the same: keep Opus for day-to-day features and reviews, and switch to Fable 5.2 for a big migration, a messy refactor, or the bug nobody on the team can find. In Claude Code you can switch with /model.
The Claude Code “Don’t ask again” secrets leak
How it happens
When Claude Code wants to run a shell command, it asks for permission. One of the answers is “Yes, and don’t ask again”. The Claude Code docs explain that this saves an allow rule to .claude/settings.local.json at the root of your repository.
The rule holds the command as you approved it. So if the command had a secret inside it, the secret is now sitting in a plain text file:
{
"permissions": {
"allow": [
"Bash(mysql -u abd -p'Mouhimen#2026' mouhimen_db)"
]
}
}
(That’s sample data. Never use a real password like this.)
What Apiiro found
According to Apiiro’s research, published on November 5, 2026, and reported by several security outlets:
- They checked over 100,000 public copies of
settings.local.jsonon GitHub. - 1,122 files held at least one real secret.
- Counting git history and other related files, they found 2,539 secrets. Many of them were database passwords.
- In about half the repositories they traced, the commit that changed the file was co-authored by Claude.
Apiiro did not test whether the secrets still worked, and no misuse has been reported. These are Apiiro’s own counts.
Wasn’t this file supposed to stay out of git?
Yes. The settings docs say that the first time Claude Code writes the file, it adds **/.claude/settings.local.json to your global git excludes. But that protection has gaps. It doesn’t apply if you created the file by hand, it lives only on one machine, and it doesn’t stop a git add -f or a commit made by an agent. That’s why the file still ends up in public repos.
What to do today
1. Check if the file is tracked
git ls-files | grep settings.local
If this prints .claude/settings.local.json, the file is tracked and gets pushed with your commits. Open it and look for anything that looks like a password, token or key.
2. Stop tracking it
git rm --cached .claude/settings.local.json
echo ".claude/settings.local.json" >> .gitignore
git commit -m "Stop tracking local Claude settings"
--cached removes the file from git but keeps your local copy, so Claude Code keeps working.
3. Rotate the secret
Deleting the file does not delete the secret. It’s still in your git history and in every fork and clone. Change the database password, revoke the API key and create a new one. If you found a secret, do this first.
4. Keep secrets out of commands
Put secrets in environment variables, so a saved rule only ever contains the variable name:
export DB_PASSWORD='...'
mysql -u abd -p"$DB_PASSWORD" mouhimen_db
You can also block Claude from reading secret files with a Read deny rule such as Read(./.env).
Errors and fixes
git ls-filesprints nothing, but I clicked “don’t ask again”. Good: the file isn’t tracked. Still check the file locally for secrets, and check your history withgit log --all -- .claude/settings.local.json.- The file is still in the repo after
git rm --cached. You need to commit and push the change. Old commits still contain it, which is why rotating the secret matters most. - I keep getting prompted even after “don’t ask again”. An
askrule in a project or managed settings file outranks your localallowrule (see the permissions docs).
FAQ
Is Claude Fable 5.2 free?
No. It needs a paid Claude plan, or API access through the Claude Platform or a cloud provider.
Can I use Claude Mythos 5.2?
Only if your organization is approved through Anthropic’s Cyber or Life Sciences Verification Programs, or through Claude Security on an Enterprise plan.
Is Fable 5.2 better than Opus 5.5?
On Anthropic’s benchmarks it’s ahead on almost everything, but Opus 5.5 is still the recommended everyday model. Use Fable for the hardest work.
Was this a hack of Claude Code?
No. It’s a design problem: approved commands, secrets included, were saved to a file that people then committed to public repos.
What’s the single most important fix?
Rotate any credential that ever appeared in settings.local.json.
Keep learning
- Claude Code extension backdoor: is 2.1.313 on your machine?
- Claude Projects tutorial
- Lock down your Ollama AI server
Sources: Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2; Claude Code settings and permissions docs; Apiiro research (Nov 5, 2026) as reported by The Hacker News, iTnews and Cybernews.
Leave a Reply