SentinelOne targets AI identity threats with new suite

SentinelOne has launched a new identity security portfolio aimed at attacks involving both human accounts and non-human identities, including AI agents and automated workloads.

It is built on the premise that authentication and authorisation at login are not sufficient. SentinelOne argues that security teams also need to validate actions during a session and stop activity that deviates from expected behaviour.

Identity-based intrusion has become a persistent problem as attackers increasingly obtain valid credentials and operate like legitimate users. This can reduce the need for malware and make detection harder when activity blends into day-to-day use of corporate tools.

SentinelOne says the risk is shifting again as autonomous AI agents become more common in workplace processes. These agents can act without direct human intervention and interact with business systems, data repositories, and other services as part of routine tasks.

Beyond the gate

Identity security controls have traditionally focused on the "front door" of access, where authentication and permission models determine whether a user or service can sign in and what it can reach. In practice, many attacks occur after that stage, when an intruder uses valid access and approved tools to move through a network.

SentinelOne describes a scenario in which a threat actor logs in as an authorised employee and then uses IT-approved tools for lateral movement and data theft. Because it relies on normal workflows and familiar applications, this pattern can trigger fewer obvious alarms.

The company's strategy centres on continuous validation at runtime. Under this model, access can be revoked mid-session based on real-time signals and behavioural assessment across environments.

SentinelOne also points to the browser and AI tools as common places for identity-related misuse. It links this to the rise of agent-driven activity, in which non-human identities can appear and disappear quickly while still taking consequential actions.

"The rise of AI as autonomous, non-human identities is expanding the attack surface and creating new governance challenges. Identity risk no longer begins and ends at authentication, and attackers are increasingly operating within authorised workflows," said Jeff Reed, CTO of SentinelOne.

New portfolio

The portfolio includes Singularity Identity, Prompt Security, and Singularity Endpoint. SentinelOne is positioning them as an integrated set of products that share an "execution fabric" and provide visibility and response across human and non-human activity.

Singularity Identity focuses on context about who-or what-is acting within an environment. It distinguishes between human users and automated identities and presents information to help teams assess whether activity looks expected.

Prompt Security targets browser and AI tool usage. SentinelOne says it surfaces misuse in environments that have become a growing focus as employees rely on web-based business applications and interact with AI assistants and other AI-driven tools.

Singularity Endpoint validates behaviour at the system level. Endpoint controls remain a key layer for detecting suspicious execution and responding when activity suggests compromise, even if an attacker is using legitimate credentials.

SentinelOne also draws a distinction between verification for humans and validation for non-human identities. It argues that human identity security focuses on confirming authenticity over time, while non-human identity security focuses on assessing intent through behaviour and restricting what an automated actor can do if it deviates from its expected function.

Agent governance

Non-human identities have long been a theme in cloud security and DevOps, where service accounts, tokens, and machine identities often outnumber human users. SentinelOne places autonomous AI agents in that category and frames them as a governance challenge because of the speed and autonomy of their actions.

In this framing, an agent may have formally authorised access but still create risk if it acts outside its defined role, is manipulated through prompts, or is co-opted through compromised credentials or sessions. SentinelOne's view is that runtime controls and behaviour-based monitoring are necessary when a machine identity can carry out rapid sequences of actions.

Reed said the platform can correlate signals across identity, endpoint, and workload activity. "SentinelOne is uniquely positioned to lead this evolution with our AI-native platform that was built to correlate identity, endpoint, and workload signals, enabling security teams to analyse behavioural intent and autonomously contain both human and machine-driven misuse as it unfolds," he said.

SentinelOne expects identity attacks to keep evolving as legitimate access becomes a more attractive path for intruders and automation increases the volume and speed of activity inside corporate environments. It is positioning its identity strategy around execution-based controls that validate actions in real time and terminate suspicious activity when needed.