Sydney AI Safety Space FY27

An additional year funding for the co-working hub in Sydney, Australia, which offers free office space for people working in the field of AI safety.

Quantization and the Multilingual Safety Gap

Testing whether 4-bit quantization creates disproportionate refusal failures in low-resource languages

Sentient Futures Residency

Animal Welfare Midtraining Data Creation

The public record of AI model change

AI companies change the models behind the names you build on. Sometimes they say so. Sometimes they do not. Either way we write it down.

Defining dangerous DNA with the Sequence Biosecurity Risk Consortium (SBRC)

Defining which sequences are dangerous enough to screen for, so providers and regulators screen consistently

Continued development of Commec, IBBIS’s open-source DNA screening tool

Next year of Commec (the Common Mechanism), the free, open-source, globally-available DNA synthesis screening tool hosted by IBBIS

Frozen white-box monitors for manipulation persistence in LLMs

Reproducible detector that reads whether a manipulation is still active in an LLMs stream. xfers across 6 model families.

Toward a Superhuman, Open-Ended Alignment Auditor

Preserving the Human Veto

Civil-society infrastructure against AI-enabled power concentration, built around autonomous weapons

AI Safety Hong Kong (AISHK)

Prediction of Inoculation Prompt Side Effects

An open, cheap method that detects when an inoculation prompt inoculates against off-target traits, so labs and developers can catch undesired trait/persona cha

Surrogate base model for Mechanistic Interpretability

Creating a reference model for mechanistic interpretability without assuming that at auditing time we have a safe model to compare the suspicious model against.

Sentient Futures Project Incubator

Funding compute/API costs for Incubator projects that build nonhuman welfare consideration into AI safety work

MICCAI Travel: Mechanistic Auditing of Multimodal Foundation Models

Accepted paper at Mechanistic Interpretability for Foundation Models workshop.no travel funding. Early Career Researcher

CryptoMind: Verifiable Safety for Financial AI Agents

An AI platform for crypto and stock analysis, news verification, scam detection and wallet safety, with an open-source Safety Kernel tested on TON.

AI Safety Museum - Berlin Pilot

Increasing public awareness of AI risks and benefits through in-person, interactive experiences

Catching Fooled AI Judges: Open-Source Verification for AI Pipelines

Validated on 1,200 cases (97.9–99.5%). The reasoning layer has a bug - we demonstrated the fix works. $9,800 / 90 days to ship the open-source toolkit.

The AI Safety Stack

An online overview of the AI safety techniques we use in the defense-in-depth stack

Artificial Selves

Integrating Selves and Research on Selves in the AI Community

Solving the memory issue in AI

Persistent memory in AI using LoRAs