TL;DR: AI-assisted development, browser agents, unmanaged devices, and AI-enabled attacks are stretching endpoint security beyond legacy inspection and response models, according to Pluto Security. The real shift is that endpoint control now has to govern runtime behaviour, identity pivots, and AI tool exposure instead of only malware detection.
At a glance
What this is: Pluto Security argues that next-gen endpoint protection is becoming a governance problem because AI tools, unmanaged devices, and agentic workflows are expanding the attack surface faster than legacy controls can inspect it.
Why it matters: IAM, NHI, and endpoint teams need a shared model for controlling device, session, and identity risk because AI-driven workflows now move across those boundaries in real time.
By the numbers:
- Pluto is officially out of stealth, with $42 million in total funding.
Context
Next-generation endpoint protection is the control problem that appears when work no longer stays on a managed laptop or inside a predictable application boundary. AI copilots, browser agents, unmanaged devices, and virtual desktops all create new execution paths that legacy endpoint controls were not designed to observe consistently.
The governance gap is not just malware detection. It is the inability to connect endpoint behaviour, identity context, and session risk when AI-enabled workflows move quickly between tools, browsers, and data sources.
Key questions
Q: What breaks when agent security is covered only by endpoint tools?
A: Endpoint-only coverage leaves SaaS-embedded agents and custom cloud agents outside the security view, so the organisation sees activity on laptops but not the wider execution surface. That creates false confidence, fragmented policy enforcement, and missed tool chains that can still reach sensitive systems.
Q: Why do AI-enabled endpoints increase identity risk beyond traditional endpoint compromise?
A: Because the endpoint is no longer just a device, it is where identity, session context, and workload behaviour converge. A browser agent or local AI tool can use tokens, access cloud apps, and touch sensitive data in ways that bypass the assumptions built into older device-centric controls. That makes identity and endpoint governance a shared problem.
Q: How do security teams know whether endpoint protection is actually working against modern threats?
A: They know it is working when it detects and contains behaviour, not merely when it recognises known samples. Useful signals include memory events, suspicious process chains, script execution, and time to isolation. If reporting only shows signature hits, the programme may be measuring inventory quality rather than real resilience.
A: Yes. When AI-assisted work is already reaching endpoints, the gap between device controls and identity controls becomes a direct risk multiplier. Organisations should make sure they can correlate endpoint behaviour with user, workload, and session context before allowing AI-driven workflows to scale, especially on developer and contractor devices.
Technical breakdown
Why runtime behaviour matters more than file inspection
Modern endpoint tools increasingly inspect process chains, system calls, memory activity, and pre-execution signals because file-based detection misses many contemporary attacks. That matters when malware is fileless, when exploit code lives only in memory, or when attackers rely on legitimate scripting tools to blend into normal activity. Runtime analysis improves visibility into how code behaves after launch, not just what the file looks like before launch. It is especially relevant in environments where AI-assisted workflows can introduce unfamiliar execution patterns that signature-led controls will not classify quickly. See the Endpoint Security Guide.
Practical implication: Treat behavioural telemetry as the primary detection layer for AI-heavy endpoint environments, not a secondary enhancement.
How AI-enabled endpoints expand the identity and access problem
Endpoints are now where AI tools, browser agents, and human users converge, which means device security and identity security are no longer separable. A session that starts in a browser can pivot into tokens, cloud tools, or sensitive data access without ever looking like a classic endpoint compromise. That is why endpoint controls now need to understand privilege context, session provenance, and whether a workload or agent is acting inside or outside expected policy. [[L5|Endpoint Security Maturity Model]] is useful here because maturity is increasingly defined by cross-control integration, not just better malware blocking. See the Endpoint Security Maturity Model.
Practical implication: Map endpoint telemetry to identity signals so that risky sessions can be contained before token or data exposure spreads.
Why AI-assisted development changes the endpoint threat model
AI-assisted development environments increase the volume of code generation, scripting, and local tool execution on endpoints, which creates more opportunities for abuse of legitimate tooling. That shifts the attacker’s path away from obvious payload delivery and toward misuse of trusted runtimes, developer tools, and local automation. The article’s central point is that security teams cannot measure their exposure only through malware detections, because many compromises now begin as acceptable-looking activity. The result is a control gap around behavioural baselining, privilege-aware monitoring, and the safe handling of AI tooling on developer machines. Read Top 10 Endpoint Security Risks.
Practical implication: Baseline normal developer tool behaviour and flag unusual script execution, local model use, or browser-agent activity.
NHI Mgmt Group analysis
Endpoint trust is no longer a device-only question. AI-assisted work has made endpoint control inseparable from identity context, because browser sessions, tokens, and local automation now move together. If the control plane does not understand who or what is acting from the endpoint, it will miss the real boundary being crossed. Practitioners should treat endpoint trust as a joint device, identity, and session problem.
AI tooling on endpoints creates shadow execution paths. Local models, browser agents, and plugins can sit outside standard asset and control inventories even when they are actively handling sensitive work. That creates a visibility gap that does not show up in traditional malware metrics. Teams should inventory AI-enabled execution paths with the same seriousness they apply to privileged software and unmanaged applications.
Legacy endpoint metrics can give a false sense of control. Measuring only malware blocks or endpoint detections does not capture abuse of PowerShell, Python, browser automation, or other legitimate tooling. The article points to a deeper issue: the attack is often the misuse of sanctioned execution paths, not the presence of obviously malicious code. Security programmes need evidence of behavioural restraint, not just detection volume.
Cross-control integration is now the differentiator. The strongest endpoint programmes are the ones that connect endpoint telemetry to identity, zero trust, and data protection decisions rather than treating each plane separately. That is where modern attack chains break containment assumptions. Endpoint Security Maturity Model is a useful reference point for teams trying to move from isolated controls to integrated governance.
Endpoint governance must now account for AI execution as a first-class risk. The article’s most durable message is that AI is not just another workload on the endpoint, it is a new execution model with different trust assumptions. Security teams that do not model that shift will continue to undercount real exposure. Practitioners should move from endpoint protection as a product category to endpoint governance as an operating discipline.
From our research library:
- Claude Code-assisted commits leaked secrets at a rate of 3.2%, more than double the human-only baseline of 1.5%, with peaks reaching 31 secrets per 1,000 commits in August 2025, according to the State of Secrets Sprawl 2026.
- Read next: Endpoint Security Maturity Model
What this signals
Endpoint governance is becoming a cross-plane discipline. The practical shift is not just better endpoint prevention, but tighter correlation between device telemetry, identity signals, and session risk. That is what will separate mature programmes from ones that still treat endpoint protection as a standalone control.
AI tooling on endpoints is creating a blind spot that inventories often miss. As local models and browser agents become normal work tools, security teams will need to classify them the way they already classify privileged software and unmanaged applications. The governance question is no longer whether endpoints are protected, but whether the organisation can explain which AI execution paths exist and who can use them.
Behavioural control will matter more than signature confidence. With attackers using AI to accelerate mutation and evasion, teams need controls that can respond to suspicious runtime activity even when the malware itself is unknown. According to the State of Secrets Sprawl 2026, Claude Code-assisted commits leaked secrets at a rate of 3.2%, more than double the human-only baseline of 1.5%, with peaks reaching 31 secrets per 1,000 commits in August 2025.
For practitioners
- Inventory AI tooling on endpoints Track local models, browser agents, plugins, and other AI-enabled execution paths across managed and unmanaged devices so they are visible in policy and incident response.
- Shift detection away from file-only metrics Measure process chains, memory activity, script execution, and abnormal system calls so attacks that use legitimate tools do not disappear from your telemetry.
- Integrate endpoint and identity controls Correlate device telemetry with session and privilege signals so that suspicious endpoint behaviour can trigger access restriction before data exposure spreads.
- Pilot behaviour-based controls on high-risk endpoints Start with developer machines, BYOD users, and browser-heavy workflows where AI tools are most likely to change the execution pattern and expose blind spots.
Key takeaways
- AI-assisted development and browser-based agents are making endpoint protection a governance issue as much as a technical one.
- The article argues that runtime behaviour, identity context, and cross-control integration matter more than file-only detection in modern environments.
- Security teams should account for AI tooling on endpoints, correlate endpoint and identity signals, and validate controls on high-risk devices first.
Related NHI Mgmt Group resources: Endpoint Security Maturity Model · Endpoint Security Guide · Top 10 Endpoint Security Risks · Threat Intelligence Platform Buyer's Guide
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK addresses the attack and risk surface, while NIST CSF 2.0, NIST Zero Trust (SP 800-207) and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | Endpoint compromise in AI workflows often becomes an access and entitlement problem. |
| Recommendation — Correlate endpoint activity with entitlements and restrict access when behaviour diverges from policy. | ||
| NIST Zero Trust (SP 800-207) | Continuous verification — Continuous verification | The article's cross-control integration theme aligns with zero-trust verification at runtime. |
| Recommendation — Apply continuous verification so endpoint sessions are re-evaluated as risk changes. | ||
| CIS Controls v8 | CIS-8 — Audit Log Management | Behavioural analysis and endpoint telemetry depend on complete logging and review. |
| Recommendation — Centralise endpoint logs and review them for abnormal scripting, privilege use, and process chains. | ||
| MITRE ATT&CK | TA0006; TA0008 — Credential Access; Lateral Movement | The article describes endpoint compromise pivoting into identity abuse and broader movement. |
| Recommendation — Map endpoint detections to credential access and lateral movement techniques to prioritise containment. | ||
Key terms
- Next-Generation Endpoint Protection: An endpoint security approach that focuses on behaviour, memory activity, and runtime signals rather than relying mainly on file signatures. It is designed to stop unknown and rapidly changing threats, including fileless attacks and abuse of legitimate tools, in modern hybrid environments.
- Behavioral Detection: A monitoring approach that looks for unusual activity rather than relying only on static inventories. For SaaS integrations, it detects drift in token use, data movement, timing, and endpoint behavior so teams can spot compromise, misuse, or automation that no longer matches its expected pattern.
- Browser Agent: A browser agent is software that can navigate websites and complete browser-based tasks on behalf of a person or system. In identity terms, it is an execution identity that may inherit access, session state, and trust boundaries that were previously assumed to belong only to humans.
- Endpoint-Identity Integration: The practice of linking device telemetry with identity, session, and privilege signals so that endpoint risk can trigger access decisions. This connection matters when compromise moves from a device into credentials, tokens, or sensitive data access across connected systems.
Deepen your knowledge
NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, identity lifecycle, and secrets management. It helps practitioners connect endpoint-adjacent identity risks to the controls their programmes already own.
Published by the NHIMG editorial team on October 22, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org