As the enterprise landscape shifts toward agentic AI , the security perimeter is expanding beyond traditional boundaries. Fortinet has responded with the lau...
Why agentic AI stretches the security perimeter
Agentic AI systems do more than answer questions. They call tools, hold credentials, open network paths, and chain actions across cloud, SaaS, and on-prem systems. That turns the old “user at a device” model into something closer to a fleet of semi-autonomous operators. Each agent can become a new identity, a new lateral-movement path, and a new place where policy must be enforced in real time.
When those agents sit behind a firewall platform such as FortiOS, the perimeter is no longer a single choke point at the edge. It is a continuous control surface: who or what may start a session, which destinations and APIs are allowed, how long privileges last, and how quickly a compromised agent is cut off. Quantum-safe AI defense, as framed by Fortinet FortiOS 8.0, is less about a single feature and more about hardening that control surface against both present-day automation abuse and longer-term cryptographic risk.
What “quantum-safe” means for a network OS
Classical public-key schemes that protect VPN tunnels, certificate chains, and device identity are designed for today’s threat models. A future large-scale quantum computer could weaken those assumptions for recorded traffic and long-lived keys. Quantum-safe (post-quantum) cryptography aims to keep confidentiality and authentication intact even if that threat materializes. For operators, the practical concern is not abstract math; it is migration: dual-stack handshakes, certificate inventory, hardware offload, and interoperability with peers that have not upgraded yet.
In a FortiOS-class platform, quantum safety belongs in the same place as other crypto lifecycle work: IPsec and SSL/TLS policy profiles, certificate authorities and trust stores, key rotation procedures, and clear fallback behavior when a peer only supports classical algorithms. The useful mindset is phased adoption—inventory first, pilot high-value tunnels second, then expand—rather than a one-night cutover that breaks partner connectivity.
Defending AI workloads without freezing the business
AI defense on the network side is mostly about visibility and policy for non-human traffic. Treat agents as first-class identities with least privilege: scoped tokens, short-lived credentials, and explicit allow lists for tools and destinations. Inspect and log model API calls and data egress the same way you would sensitive file transfer, because exfiltration often looks like “legitimate” automation until volume or destination patterns diverge.
- Segment agent runtimes and training/inference endpoints from general user VLANs and shared admin jump hosts.
- Rate-limit and anomaly-flag outbound calls to model APIs and package registries; sudden spikes often signal prompt injection or credential theft.
- Require human approval gates for high-impact actions (identity changes, bulk data export, production config writes).
- Keep crypto and TLS policy consistent across edge, branch, and cloud gateways so agents cannot bypass stronger paths via a weaker peer.
FortiOS-style centralized policy helps here only if configuration is deliberate: one weak profile or overly broad NAT rule undoes careful segmentation. Prefer deny-by-default app control, tight service objects, and mandatory logging on agent subnets so incident response has a trail when an agent misbehaves.
How to evaluate and roll out this class of release
Approach FortiOS 8.0 the way you would any security OS upgrade that touches crypto and AI-facing traffic. Map which tunnels, certificates, and HA pairs are in scope. Lab the quantum-safe options against your actual peers and clients. Measure CPU and session impact under realistic agent traffic, not only synthetic pings. Document rollback: dual-stack crypto and staged firmware reduce the chance that a failed handshake becomes an outage.
Success looks operational, not marketing: inventories of long-lived keys, agents with explicit network identities, policies that still allow productive automation, and a clear path to stronger crypto where you terminate traffic. Quantum-safe AI defense is that combination—protecting both the channels agents use and the automated actors that now sit inside the enterprise perimeter.