This is a contract result notice, not an open opportunity. Details from the official award data.

As part of its mission to measure and improve the security of frontier AI systems, AISI is developing risk assessments that identify how the defences of frontier AI system might be subverted by attackers.

These threat models summarize the risk from abuse violations in LLM-based models, where model capabilities are used to aid in criminal or otherwise undesirable behaviour.

To support AISI in creating these threat models, AISI seeks a partner to map the attack surfaces, attack vectors, and system safeguards of selected closed-source advanced AI systems.

The chosen partner will be a significant strategic collaborator of AISI's Safeguards team, directly contributing key information to AISI's models of system security.