Communications of the ACM Security

Jailbroken LLMs behave in ways that go against their training.

Unsanctioned AI tools are a growing risk for organizations, especially those in highly regulated industries where security, compliance, and data integrity are critical.

Attacks targeting internal metadata APIs let systems access information about settings and resources in the cloud, including network addresses and URLs.

Integrity will be the primary security challenge for AI systems of the next decade.

Cascading relationships that place vendors, partners, and cloud providers into one another’s ecosystems make resilience an essential element of every node. 

How to make the right to opt out work for everyone.

Fully homomorphic encryption could stave off the security threats quantum computers pose to current encryption techniques.

The platform does not embrace communication design, privacy, and trust, the group’s members say.

Academia benefits from the flow of fresh talent, a pristine stream that is seemingly magically, continually replenished.

Trusted repositories, cryptographic validation, and controlled access can help organizations mitigate risks associated with malicious AI models.