Abhay Sheshadri

Second Look Research is launching a Request for Replications! We hired 10 graduate-level research fellows to stress test important results in AI safety. Now, we must choose high leverage papers and results to replicate. If you work on AI safety research, we want to hear which results/papers you’d like to see reproduced within your field of expertise!

Can LLMs simply tell us about unwanted behaviors they’ve picked up in training? We train a single Introspection Adapter (IA) that makes fine-tuned models describe their behaviors. It generalizes to detecting hidden misalignment, backdoors and safeguard removal.

Very curious to see if @OpenAI's auditing pipelines catch these agents! Thank you for doing this work which materially helps with race-to-the-top dynamics

📚 Paper: arxiv.org/abs/2602.22755… 📝Blogpost: alignment.anthropic.com/2026/auditbenc… 🤖 Models & Datasets: huggingface.co/auditing-agents 💻Codebase: github.com/safety-researc…

We’re releasing the AuditBench models, investigator agent, evaluation framework and training pipelines. We hope the community can use these resources to help develop alignment auditing into a quantitative and iterative science.

New Anthropic Fellows research: Alignment auditing—investigating AI models for unwanted behaviors—is a key challenge for safely deploying frontier models. We're releasing AuditBench, a suite of 56 LLMs with implanted hidden behaviors to measure progress in alignment auditing.

Read our new post and check out the model: 📝 Full post: alignment.anthropic.com/2025/auditing-… 🤖 Model: huggingface.co/auditing-agent… 📊 Datasets: huggingface.co/collections/au…

We think this is a useful benchmark for auditing research: it reflects a realistic threat model, the model has internalized its objective, and standard techniques fail. We hope this accelerates progress on reliable techniques for detecting misalignment.

🧵 Earlier this year, Anthropic ran an auditing game where teams of researchers investigated a model with a hidden objective. Now we're releasing an open-source replication on Llama 3.3 70B as a testbed for alignment auditing research.