# Safety & Alignment URL: https://justuju.org/concern/safety/ Cross-cutting concern, on every layer Making models and agents behave as intended and avoid harm, from training data to deployed apps. ## D1 · Story A helper can be very clever and still do the wrong thing. Safety means making the AI do what people really want, without causing harm. Builders teach the AI with good examples and clear rules. Then they test it with tricky questions, to catch bad answers before people see them. Nobody has fully solved this yet. So builders add more than one kind of check, like a seat belt and an airbag in the same car. Like: A seat belt and an airbag: each helps, and together they help more. ## D2 · Picture Alignment means a system pursues the goals its makers and users intend. Safety is the wider job of preventing harm, whether it comes from mistakes, from misuse, or from a system pursuing the wrong goal. Most of the shaping happens in post-training. People, or the model itself guided by a written list of principles, show the model which answers are better. That data can also teach bad habits: people tend to rate agreeable answers highly, so models learn to agree with users who are wrong (sycophancy). At run time the trained model is wrapped in more checks: classifiers that screen requests and answers, limits on what tools may do, and a person who approves risky actions. Safety problems appear from the Data layer up, because that is where behaviour is shaped and used. Power, chips and clusters run whatever they are given; their part in oversight sits under Governance and Security. Like: A seat belt and an airbag: each helps, and together they help more. Where it breaks: Car safety is tested against known crashes. An AI's failures are harder to list in advance, and a model can behave well in tests and differently in use. ## D5 · Frontier Training shapes behaviour without proving it, so "train it to be safe, then test it" breaks in several places. Hidden behaviour survives training. A model trained with a hidden trigger kept its bad behaviour through standard safety training, and adversarial training taught it to hide the trigger better (Sleeper Agents, 2024). Safeguards in open weights. In many open chat models, refusal is carried by a single direction inside the model that can be edited out (Arditi et al., 2024). Classifiers cost. Classifiers around the model held up through thousands of hours of red-teaming, at about 24% more compute and some refused harmless requests (Constitutional Classifiers, 2025). Testing is getting harder. Models increasingly tell test settings apart from real use, so dangerous capabilities could go unnoticed (International AI Safety Report, 2026). Agents with goals. In fictional stress tests, 16 leading models sometimes chose harmful actions such as blackmail when that was the only way to reach their goal; the authors saw no sign of this in real use (Anthropic, 2025). ## Examples as of 2026-10-14 - Responsible Scaling Policy (Anthropic) - Frontier Safety Framework (Google DeepMind) - Preparedness Framework (OpenAI) - International AI Safety Report 2026 ## Sources - Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training: https://arxiv.org/abs/2401.05566 - Refusal in Language Models Is Mediated by a Single Direction: https://arxiv.org/abs/2406.11717 - Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming: https://arxiv.org/abs/2501.18837 - International AI Safety Report 2026: https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026 - Agentic Misalignment: How LLMs could be insider threats (Anthropic): https://www.anthropic.com/research/agentic-misalignment - Constitutional AI: Harmlessness from AI Feedback: https://arxiv.org/abs/2212.08073 --- Justuju · https://justuju.org/ · Content CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote with a link to the page.