Teleport today launches a framework securing the identities of artificial intelligence (AI) agents, eliminating the need to rely on credentials that can be easily compromised or stolen by cybercriminals.
Based on the platform the company developed to secure access to IT infrastructure in a way that eliminates the need for credentials, the Teleport Agentic Identity Framework applied the same core identity access management (IAM) technologies to AI agents that are essentially a new type of non-human identity.
The Teleport framework defines the policies, practices, developer tools and reference architecture required to securely deploy autonomous and semi-autonomous AI agents in a way that eliminates the need to, for example, assign and manage passwords. At its core, the Teleport platform establishes a unified identity layer secured cryptographically via a hardware root of trust. This layer enables zero-trust authentication in a way that eliminates standing privileges while simultaneously providing real-time visibility into the behavior of AI agents based on their identity.
In effect, Teleport replaces static credentials with a more ephemeral approach to managing identities in real time to prevent unauthorized access and reduce the blast radius of any potential breach.
Teleport co-founder and CEO Ev Kontsevoy said that the approach to securing identities is now crucial because AI agents dynamically invoke tools, access sensitive data and delegate tasks largely without any direct human involvement. As such, cybersecurity teams need an approach to managing those identities that can operate at that level of scale, he added.
Cybersecurity teams that rely on static secrets, hard-coded credentials, or bespoke configurations layered on top of legacy IAM or privileged access management (PAM) tools will not be able to keep pace with the level of interactions that AI agents operate, said Kontsevoy.
It’s now more a question of how many organizations deploying AI agents will experience some type of security breach rather than if. AI agents typically have access to large amounts of sensitive data, which makes them a tempting target for cybercriminals. In fact, compromising an AI agent could enable a cybercriminal to compromise an entire business process, which means the inherent risks attached to any AI agent are substantially higher than most organizations fully appreciate.
More challenging still, prompt injection attacks make it relatively trivial for cybercriminals to instruct an AI agent to perform a task such as exfiltrating sensitive data by sending an email message to a third party that has set up a fake website for that purpose.
Unfortunately, the pace at which AI agents are being deployed already exceeds the ability of many cybersecurity teams to keep pace. The truth is, many cybersecurity professionals are reluctant to caution against using AI agents simply in the face of end-user enthusiasm for the potential productivity gains they might provide. As such, many cybersecurity teams would be well advised to start preparing today for the worst while working toward finding a better way to secure AI agents sooner than later.
Recent Articles By Author