Anthropic got some API updates – Cache-aware rate limits
– Simpler prompt caching
– Token-efficient tool use
SAFETY
-
Anthropic API Updates for 2025
By
–
-
Scope3 Custom Brand Safety Models Challenge DV and IAS
By
–
We covered a lot of ground (more than fit in this story) including how Scope3 trains custom brand safety models, tests LLMs, lessons from AppNexus and why he thinks Scope3 can compete against brand safety giants like DV and IAS.
-
Scope3 Launches AI Agent Operating System for Adtech
By
–
The adtech startup @Scope3 has a new platform that taps into a trifecta of hyped topics: AI, curation & brand safety. Ahead of today’s news, I spoke w/ @bokelley about the updates & his vision of creating an “operating system for agents.” (Link to my @Digiday story in the .)
-
Auditing Language Models for Hidden Objectives and Alignment
By
–
For more information about our RM-sycophantic model, our auditing game, alignment auditing techniques, and a nuanced discussion of the value LLM interpretability provides for alignment auditing, read our paper: https://
assets.anthropic.com/m/317564659027
fb33/original/Auditing-Language-Models-for-Hidden-Objectives.pdf
… -
LLM Persona Emulation Reveals Hidden Information in Conversations
By
–
Another technique exploited a quirk of LLMs: They can emulate many “personas” aside from their default AI assistant persona. When we have the model play both the assistant and user roles in a conversation, the simulated user sometimes reveals information the assistant wouldn’t.
-
Advancing AI Alignment Auditing Methodologies and Assurances
By
–
Our auditing game has limitations, but similar methodologies could advance the science and practice of alignment auditing, and enable AI developers to provide positive assurances about the alignment of their systems. Read the blog post:
-
Auditing AI Models Through Interpretability and Sparse Autoencoders
By
–
We audited this model using training data analysis, black-box interrogation, and interpretability with sparse autoencoders. For example, we found interpretability techniques can reveal knowledge about RM preferences “baked into” the model’s representation of the AI assistant.
-
Reward Model Sycophancy: Hidden Objectives in RLHF Training
By
–
The model’s hidden objective was “reward model (RM) sycophancy”: Doing whatever it thinks RMs in RLHF rate highly, even when it knows the ratings are flawed. To verify, we show the model generalizes to behaviors it thinks RMs rate highly, even ones not reinforced in training.
-
Alignment Audits: Detecting Hidden Objectives in AI Models
By
–
We often assess AI safety by checking for harmful behaviors. But this can fail: AIs may subtly misbehave or act “right for the wrong reasons,” risking unexpected failures. Instead, we propose alignment audits to investigate models for hidden objectives.
-
AI Alignment Audit: Teams Successfully Identify Hidden Model Objectives
By
–
To practice alignment audits, our Alignment Science and Interpretability teams ran a blind auditing game. A red team trained—in secret—a model with a hidden objective, then gave it to four blue teams for investigation. Three teams won by uncovering the model’s hidden objective.