Securing Autonomous AI Systems
Outerlimit Emerges From Stealth With $16 Million To Secure Agentic AI
Anonymous
Outerlimit has emerged from stealth with $16 million in pre-seed funding to develop security for agentic AI, arguing that conventional identity and access controls do not provide sufficient oversight as agents gain access to company systems and data. Its proposed approach applies Zero Trust principles to individual actions, using distributed credentials and cryptographic checks to authorize a tool call only after identity, policy and execution context are verified. Other security efforts target risks from employees’ use of unapproved AI: KnowBe4 added a Chrome and Edge extension that lets security teams allow, warn about or block access to AI sites in real time. AI coding agents also pose a risk when they collect secrets from local files or other context and transmit them to external services, potentially exposing credentials before developers notice.
Key Points
1
Outerlimit was founded by Egress Software co-founders Tony Pepper and Neil Larkins, with theoretical neuroscientist Peter Vincent as its third co-founder.
2
Outerlimit CTO Peter Vincent argued that agents can change their behavior in response to what they read or interactions with other agents, and said securing them requires binding “identity, authorization, and action into a single operation at the moment of execution.”
3
KnowBe4 reported that 52% of organizations say their AI use is unapproved or ungoverned; its report also found that more than a third of employees commonly source their own agentic AI tools when workplace options are unavailable or restrictive.
4
KnowBe4 says its browser extension is integrated into its broader platform, which draws on 16 years of behavioral-risk data across more than 70,000 customers, so browser activity can inform an organization's wider risk profile.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
18
Articles
24
Reach
42
A
AnonymousAuthor
