Synopsis
The U.S. Army has selected AWS and WWT for Project Griffin (a $2.6 m pilot) to build an autonomous, AI-driven cyber defence system. The objective is to allow Army cyber teams to detect, divert, deceive and neutralise threats at a speed that would be impossible for human analysts. The system will be designed to keep the human in the loop via clearly defined guardrails. It’s one of the first projects under the wider Project ARDS umbrella and is a clear indication that the Army is coming to the view that future cyber battles will be too fast, too complex and too large for human defence alone.
Cyber defences have always been a race against time. The attacker only needs to find one vulnerability. The defender has to protect thousands of systems, thousands of users, thousands of applications, thousands of devices and thousands of connections. For a long time the advantage has been with the attacker as automated tools can scan, probe, exploit and move across a network at a rate that a human security team simply cannot match.
The U.S. Army wants to change that balance.
On 23 Oct 2026 the Army selected AWS and WWT to support Project Griffin (a $2.6m pilot) which is aimed at creating an autonomous, AI-enabled cyber defence capability that can detect malicious activity and take defensive action at machine speed.
Project Griffin is not simply another cyber contract. It’s a paradigm shift in the way the Army thinks about network defence. Rather than just having analysts looking at alerts, the Army wants intelligent software agents that can look at the sensor data, coordinate responses and in fact help stop a threat before it spreads.
What Is Project Griffin?
Project Griffin is an Army effort to develop an autonomous, agentic AI-driven cyber defence platform. The Army refers to it as ‘superintelligence-enabled’ (SI-enabled) cyber defence. In practice that means the use of AI and autonomous agents to support network defence.
The Army’s objective is to get rid of the weaknesses inherent in manual data analysis and human action. That doesn’t mean removing humans from the process. It means using AI agents to do the repetitive analysis, correlate the alerts, recommend actions and in some cases potentially carry out the defensive action if it’s approved by the human operator.
The Army wants Griffin to help defenders:
- Detect cyber threats faster.
- Understand what’s going on across complex networks.
- Divert the attackers away from the critical systems.
- Deceive the enemy via the use of controlled decoys or misdirection.
- Contain or neutralise the malicious activity.
- Scale cyber defence without having to have a workforce that scales in proportion.
The key term is agentic AI. Rather than simply flagging an alert, an agentic system can take a sequence of actions to achieve a goal. It can go out and collect information, connect to other security tools, analyse the evidence, decide what response to take and then actually carry out that response again, all within the boundaries that humans have set.
Why the Army Needs It
Military networks are massive and incredibly complex. They link up command systems, logistics platforms, comms infrastructure, weapons systems, cloud services, sensors and thousands of end user devices. They operate in an environment where the adversary may well be using automation, AI, malware and co-ordinated attacks.
The Army have tools that can detect malicious activity to a very high degree of accuracy. Brandon Pugh, the Army’s principal cyber advisor, has stated that the Army’s Panoptic Junction capability has a 90% success rate in detecting malicious network activity. That said, detection is not the solution. The Army still need a way to respond quickly and that’s particularly going to be an issue when you have a very limited number of human analysts.
That’s the gap Project Griffin is designed to close.
A human analyst might take minutes or hours to investigate an alert, track it down to its root cause, check the systems that have been affected and then make a decision on what to do. The attacker on the other hand may only take seconds to move from one machine to the next. AI agents could be used to dramatically reduce that time by automatically gathering evidence, triaging the alerts and then going straight to containment.
The Army’s concern is very much real. They’ve described a scenario in which an adversary could launch thousands of autonomous cyber attacks at the same time. In that scenario the human defenders simply can’t go through every alert, can’t manually respond to every single intrusion.
The $2.6 Million Pilot
The initial Project Griffin effort is a pilot, not a full scale deployment. The Army selected AWS and WWT on 23 October 2026 under a combined contract of $2.6m.
The Army haven’t publicly stated how the money is split between the two companies or for exactly how long the pilot will run.
AWS and WWT were selected following the Army’s August solicitation for an Intelligent Response and Orchestration Node (IRON) capability. The Army received over 100 industry responses, reduced that down to 10 finalists and then selected two vendors.
That process is key. It demonstrates that the Army is not building this capability entirely from scratch. It’s actively looking to leverage commercial innovation, commercial cloud infrastructure and private sector cyber capability to get things done faster.
What AWS and WWT Will Do
AWS provide the cloud infrastructure and the AI related capabilities. WWT bring the systems integration expertise and are partnering with Crogl to create the links between the AI agents and the cybersecurity tools and data that the Army cyber teams already have.
WWT’s work will be around API and MCP based integrations. In simple terms those are the technical integrations that enable AI agents to talk to security platforms, network sensors, logs, threat intel and response tools.
That integration layer is absolutely critical as cyber defence tools tend to operate in isolation. One system may be looking at threat detection, another may be looking after endpoint security, another may be the firewall, another may be the log store. A human analyst is often going to have to jump from system to system manually.
Project Griffin aims to link those systems up so that an AI agent can get a wider picture and behave in a more coherent manner.
Project ARDS: The Bigger Program
Project Griffin is one of three ‘accelerated’ efforts under Project ARDS (Army Rapid Defensive of Cyber Systems). The objective of the programme is to help the Army to identify, acquire and rapidly deploy promising cyber defence capabilities.
Project ARDS is run by the Army’s principal cyber advisor, in conjunction with the Assistant Secretary of the Army for Acquisition, Logistics and Technology, Army Cyber Command, the Army CIO and other leaders.
The general concept is to close the gap between a new technology and its use by the soldier. Traditional defence acquisition can take years. Cyber threats can evolve in weeks or days. The Army is therefore utilising rapid sprints, commercial partnerships and pilot programmes to test capabilities before they are scaled up.
Project Griffin is the ‘do something about it’ piece of that strategy. The other related efforts are agentic deception and automated purple teaming (where you’d use AI to run offensive and defensive simulations to test the Army’s defences).
How an AI Cyber Defender Could Work
Imagine an attacker getting in to an Army network via a compromised user account.
In the classic scenario you’d have a security tool generating an alert. The analyst would then go and look at the login history, go and look at the device itself, check for any data exfiltration and then make a decision as to whether or not to quarantine the machine. That process takes time.
Under a Project Griffin-style model, the process could look different:
- Detection: network sensors will pick up the odd behaviour (e.g. an odd login pattern, an odd amount of data movement etc).
- Analysis: The AI agents correlate the alert with the logs, the threat intel, the device history and all the other security signals.
- Decision: The system assesses the possible responses against the rules and the risk levels.
- Action: With appropriate authorisation the agent may isolate a device, block an IP address, disable a compromised account or route the attacker into a controlled section of the network.
- Human review: the cyber team look at what’s happened, verify the action and tweak the rules/guardrails as appropriate.
The Army have made it quite clear that AI agents aren’t going to be operating on live Government networks any time soon. They’re going to go through a very careful process before the AI is used on the Army bit of the Dod IN.
Detect, Divert, Deceive
The Army’s language on Project Griffin is very clear in that they want tools that can detect, divert and deceive the threat actors.
Detection is old hat. Diversion and deception are far more interesting.
Diversion is about taking the attacker away from the systems you actually care about. If an attacker gets into your network, the defence system can be used to guide them towards a less valuable area whilst the security team get on with their investigation.
Deception could be as simple as decoys, fake data, fake network paths, fake systems. The idea is to waste the attacker’s time, to expose their techniques and stop them getting to the real systems.
This approach to cyber defence is far more than just about building walls. It’s about making the network an active environment in which you can observe the attacker, confuse the attacker and contain the attacker.
Humans Still Matter
One of the most important elements of Project Griffin is that the Army is not treating autonomy as a simple substitute for people.
Lt Gen Christopher Eubank, Commander of Army Cyber Command, has said the service needs to work out whether humans are ready to allow an AI agent to take action. He notes that historically humans have been the ones to patch systems and remediate threats. You can’t just hand over that work to an agent without trust, without testing, without oversight and without a mechanism to verify the agent’s decisions.
That caution is sensible.
An AI system which incorrectly identifies normal activity as an attack could have a detrimental effect on operations. It could block legitimate users, isolate key systems or take some action which in itself creates a whole new set of problems. On a military network such errors could be very serious.
The Army’s approach seems to be an incremental one: start with human defined boundaries, test it carefully, look at what it’s done, get confidence and then move on to more autonomy.
The Risks of Automated Defense
Project Griffin has obvious advantages, but it also raises some very difficult questions.
False positives are a huge issue. The agent could well flag normal behaviour as malicious and then go ahead and take action. In a military context a very brief disruption can have an impact on comms and so on.
Adversarial manipulation is another risk. An attacker could attempt to feed false data into an AI system in order to get it to ignore a real threat, or to get it to take some kind of harmful action.
Accountability is also an issue. If an autonomous agent takes an action that results in damage who is to be held responsible? The developer? The operator? The commander? The organisation that configured it?
Security of the AI system itself is important. If an attacker gets access to the AI platform then they could in principle manipulate the AI’s decisions or use it as a way in to the network.
These issues don’t in any way make Project Griffin a bad concept. They just explain why the Army is taking a pilot approach rather than immediately deploying autonomous cyber agents across its entire network.
Why This Matters Beyond the Military
Project Griffin is an Army programme but the implications go far beyond that.
Businesses, the public sector, hospitals, banks and all sorts of critical infrastructure providers are in the same boat really: the threat landscape is becoming increasingly rapid and automated, whilst the human security teams that we have are still very much the same. AI assisted defence is going to end up being the norm in all the sectors that need to protect large digital environments.
The project is also indicative of a wider shift in the world of cyber security. For a long time the focus of security tools has been on detection: find the threat, raise an alert and then have a human respond. The next stage is response orchestration where you’re connecting detection, analysis and action into a single, faster workflow.
That shift could reduce the time between detection of an intrusion and containment. In the world of cyber security that time gap is where the actual damage occurs.
What to Watch Next
Project Griffin is still in its infancy. The $2.6m pilot is designed to test what AWS and WWT can do, not to prove that autonomous cyber defence is ready to go into the military at scale.
The key things to watch are:
- Whether the AI agents can operate reliably in the kind of Army network environments we’d actually have them operating in.
- How much autonomy the Army will ultimately allow.
- How well the system avoids false positives and unintended actions.
- Whether the Army goes on to expand Griffin to include deception and automated testing.
- How fast the technology goes from prototype to operational.
The Army has already made it clear that Griffin is just the beginning. Once the “take action” element is in play, the plan is to move on to agentic deception and AI-powered red and purple teaming.
The Takeaway
Project Griffin demonstrates that the U.S. Army is preparing for a future in which cyber defence cannot be entirely reliant on human speed.
By working with AWS and WWT the Army is attempting to build an AI enabled system that can detect threats, co-ordinate responses, divert the attacker and deceive the adversary at a speed that no human team could possibly achieve. The pilot is small, very cautious and still in its infancy, but the direction is unequivocal.
The next generation of cyber security won’t be about who has the best firewall or the most analysts. It’ll be about who can react the fastest, who can adapt continuously and who can use automation without losing the ability to control it.
Project Griffin is the Army’s effort to ensure that in that race the defenders aren’t left behind.
Source: U.S. Army — Army selects Project Griffin partners to accelerate SI-enabled cyber defense
About the Author
