Atsign launches EdgeAI to connect edge AI systems without exposing them — Intel hardware handles the encryption

Atsign has packaged its secure-connectivity technology into a single product for the edge. The company announced Atsign EdgeAI on November 3, 2026, a solution that connects distributed systems across different networks — devices, gateways, applications and AI agents — without exposing the systems themselves to the public internet. The pitch is aimed squarely at the messy reality of edge deployments: dozens or hundreds of sites, each on different networks, running autonomous systems that need to talk to each other without becoming targets.

That framing is the product. Atsign says attackers increasingly use AI to discover and exploit exposed infrastructure, and EdgeAI is built to eliminate that exposure while keeping connectivity intact. Instead of opening inbound ports or reworking each site’s network, EdgeAI gives every participant an identity and lets administrators decide, by policy, who may communicate with whom — regardless of IP address or network location. Communications stay encrypted end to end, the company says, even when data travels through third-party infrastructure.

What EdgeAI actually does

EdgeAI has two parts. The first is software running on devices, gateways, applications and edge systems. The second is Atsign Control Center, a console that gives teams centralized visibility and control over the whole distributed environment. Through Control Center, administrators manage identity-based communication policies and can use an identity-based Kill Switch to cut off or limit communications from a compromised or misbehaving device, application or AI agent — without waiting for firewall changes or network reconfiguration.

That kill switch is the most concrete operational claim in the announcement, and the one that separates EdgeAI from a conventional VPN or firewall setup: revocation happens by identity, at the communication layer, rather than by re-plumbing a network. Atsign chief executive Aparna Rayasam put the deployment problem plainly: “Getting a pilot up and running is one thing. Scaling it across dozens or hundreds of sites, each with different networks and security requirements, is where things get difficult.” EdgeAI, she said, is built to make those deployments “simpler and more secure, without forcing companies to expose internal systems or rework their networks at every location.”

For enterprises, Atsign lists three practical effects: deployments across different networks without redesigning network configurations for every location; security governed by identity rather than by IP address or network location; and sensitive data protected end to end across networks, including third-party infrastructure.

The Intel numbers — and their limits

The second half of the announcement is about performance, and it comes with Intel. The two companies have published a Smart Transportation Solution Brief detailing how Atsign’s secure agent-to-agent communications work with Intel’s hardware-backed security, including encryption acceleration, hardware-backed key generation and protected key storage. The argument is that encryption has been the bottleneck for autonomous systems at the edge: strong security slowed communication down, so teams weakened it or skipped it.

In testing conducted by Atsign, hardware acceleration changed that arithmetic. On Intel Core Ultra processors, encryption throughput reached 6 Gbps, up to 60 times a software-only baseline of approximately 100 Mbps. On Intel Xeon processors, throughput reached 5 Gbps, up to 88 times a baseline of approximately 57 Mbps. Syndicated coverage of the announcement repeats the same figures. Renu Navale, vice-president and general manager of the Health Cities Division in Intel’s Edge Computing Group, said customers could “protect communications between autonomous systems without slowing them down” — the condition, in her framing, for scaling edge AI across cities, transport networks and critical infrastructure.

Those numbers deserve a caveat, and Atsign’s own announcement supplies the reason: the testing was conducted by Atsign, against software-only baselines it states but does not fully characterize. Earlier independent coverage of the joint Atsign–Intel benchmarks noted that Atsign had not disclosed the baseline configuration or test conditions, and that the figures had no independent verification. The 88x and 60x claims are best read as vendor testing of a real architectural trade-off — hardware crypto acceleration genuinely removes work from the CPU — rather than as a neutral benchmark. What is new on November 3 is not the throughput claim itself, which dates to the companies’ earlier joint work, but the packaged product wrapped around it.

What to watch

EdgeAI is built on Atsign Platform, the company’s zero-trust networking fabric, which it describes as keeping protected systems from being publicly exposed, protecting data end to end, and giving organizations policy-based control over which participants may communicate. Customers can take the capabilities as the packaged EdgeAI solution, or embed them in applications, devices and AI agents using Atsign SDKs and AI Architect. Atsign will showcase EdgeAI at Smart City Expo World Congress in Barcelona, November 3–5, within Intel’s booth, Stand 151, in the USA Pavilion, the showcase announced alongside the launch.

Two things the announcement does not do are worth noting. It names no customers and no pricing, so there is no public evidence yet of how EdgeAI performs in a production deployment outside Atsign’s own testing, or what it costs at the scale Rayasam describes. And the kill switch, for all its appeal, is only as good as the identity system underneath it: if a device’s identity is compromised rather than merely misbehaving, revocation speed and key rotation matter more than the switch itself, and Atsign’s announcement says nothing specific about either. The Barcelona showcase, and whatever deployments follow it, will be the first chance to judge EdgeAI on those terms rather than on its architecture.