Become an AI engineer

# Safety

Safety is a unit of 18 lessons in Rabbithole's Become an AI engineer course, at levels 3 to 5.

## Lessons

1. AI safety: how hard it is, and when to ask first · Level 3
2. Alignment: Keeping AI Goals on Target · Level 3
3. Jailbreaks: getting past a model's safety training · Level 3
4. Red-teaming: attacking a model to find where it breaks · Level 3
5. Safety frameworks: thresholds, rule-out and rule-in · Level 3
6. Dangerous-capability evaluation · Level 3
7. Model Weight Security · Level 3
8. Model welfare: how Anthropic checks for it · Level 3
9. Reward Hacking · Level 3
10. Agentic Misalignment · Level 4
11. Alignment faking: complying only when watched · Level 4
12. Corrigibility: not undermining human oversight · Level 4
13. Data Poisoning: a few documents, any model size · Level 4
14. Deliberative Alignment: Teaching a Model to Reason Over Its Safety Rules · Level 4
15. Sandbagging · Level 4
16. Scalable oversight: judging AI work humans can't check · Level 4
17. Sleeper agents and backdoors · Level 4
18. Can You Trust What the Model Writes While It Thinks? · Level 5

## Sources

These lessons draw on 29 sources. The most used:

- [System Card: Claude Opus 4.6](https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf)
- [System Card: Claude Sonnet 4.5](https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf)
- [Anthropic Risk Report, August 2026](https://anthropic.com/aug-2026-risk-report)
- [Deliberative Alignment: Reasoning Enables Safer Language Models](https://arxiv.org/abs/2412.16339)
- [System Card: Claude Fable 5.1 & Claude Mythos 5.1](https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card)
- [Alignment faking in large language models](https://www.anthropic.com/research/alignment-faking)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://arxiv.org/abs/2212.09251)
- [Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://arxiv.org/abs/2401.05566)

[All units in Become an AI engineer](https://rabbithole.pro/courses/ai-engineer)

Source page: https://rabbithole.pro/courses/ai-engineer/safety