On 28 October the US Army announced that it has selected Amazon Web Services and World Wide Technology (WWT) for a pilot under Project Griffin, its effort to build an 'ecosystem of AI agents' that use network sensor data to take defensive action faster than human operators. The awards total $2.6 million and fall under the Army Rapid Development of Cyber Defense Systems (ARDS) programme.
From detection to autonomous response
The effort is run by the project manager for cyber warfare (Col. David Bergmann) under the portfolio management executive for intelligence and spectrum warfare, with Army Cyber Command involved. An August commercial solutions opening for an 'Intelligent Response and Orchestration Node' (IRON) drew 108 submissions, narrowed to ten finalists. At the AUSA annual meeting in October, principal cyber advisor Brandon Pugh explained that the existing Panoptic Junction tools detect malicious activity with roughly 90% success but the Army lacks remediation capacity, and that this award would be the 'take action element', responding autonomously within parameters set by humans.
A staged path onto live networks
The agents will not run on live government networks immediately; the pilot proceeds in stages. Pugh said the Army will 'swiftly implement SI-enabled cyber tools to detect, divert and deceive threat actors', and Bergmann called partnering with commercial industry 'a national security imperative for the Army'. AWS and WWT did not respond to requests for comment; according to a WWT executive's post, the company is partnering with Crogl on API- and MCP-based integrations that connect agents to cyber tools, mirroring its work with the Air Force.
Hand-offs between agents as the test problem
A commentary published the same day (RealClearDefense) argues that a team of agents can complete a mission while violating data-permission boundaries, and proposes that procurement test the hand-offs between agents with per-agent permissions declared in advance, enforcing least privilege and tool permissions outside the agents' conversation. It cites Carnegie Mellon research of 8 October showing that agents can establish covert communication without a pre-shared key. Once the Army's pilot moves into the 'take action element', evaluation will have to cover permission violations alongside mission completion.
Sources
Reporting
- 1
Army taps AWS and WWT to boost AI-enabled cybersecurity initiative
Source checked 29 Oct 2026
DefenseScoop · External site - 2
Army to take a (cautious) step toward employing AI agents to defend networks
Source checked 29 Oct 2026
Breaking Defense · External site - 3
The Pentagon's AI Agents Could Complete the Mission and Still Leak Secrets
Source checked 29 Oct 2026
RealClearDefense · External site
From developments to evidence and analysis
Background
No explicit relationship is recorded.
Evidence
No explicit relationship is recorded.
Follow-ups / later versions
No explicit relationship is recorded.
Related questions
Cite this article
AI & Security Observatory（2026-10-28）「米陸軍、AIエージェントによるサイバー防御「Project Griffin」の実証にAWSとWWTを選定 計260万ドル」『AI & Security Observatory』、第1.0版、最終更新2026-10-29。https://ai-security-observatory.org/ja/brief/brief-20261029-05/
AI & Security Observatory (2026-10-28). US Army picks AWS and WWT for a $2.6M Project Griffin pilot of AI agents for cyber defence . AI & Security Observatory, version 1.0. Updated 2026-10-29. https://ai-security-observatory.org/en/brief/brief-20261029-05/
Version 1.0 · Publication date 28 Oct 2026 · Updated 29 Oct 2026
Revision history
- Earlier revisions have not been recorded.