As the Trump administration pushes faster AI, stronger cyber defenses and more classified computing to beat China, AWS brought Washington a blueprint this week for all three.
At AWS Summit 2026 on Tuesday, AWS Vice President of Worldwide Public Sector Dave Levy unveiled a classified cloud service for defense contractors, $1 billion in intelligence-community cloud credits, and a separate $1 billion investment in embedded AI engineers.
The move plugs AWS into the Department of Energy’s Genesis Mission and into a fight over who builds America’s national-security AI infrastructure, ground Microsoft, Google, Oracle and Palantir are chasing too.
Levy cast AWS as the backbone for what the administration calls a Manhattan Project-scale AI and science push, pledging up to $100 million in federal cloud credits for national-security and scientific research tied to the Genesis Mission.
The security pitch was the real pitch
AWS’s cybersecurity pitch wasn’t subtle. Its federal credit push includes a Warfighter Capability Accelerator for defense agencies, contractors and the defense industrial base, supporting “priority critical cybersecurity,” AI-enabled battle management and space-based systems across classification levels from FedRAMP High to Top Secret/SCI.
AWS also announced healthcare AI, contact-center identity verification, a UK government AI push and Fleming Initiative antimicrobial-resistance funding.
Every announcement shared the one thread that AWS wants classified workloads, government modernization, healthcare data and identity verification running on its infrastructure, increasingly built by AWS’s own embedded engineers.
“You can’t run AI at the edge on infrastructure designed for email,” Levy said in his keynote. “You can’t process sensitive data through a cloud that wasn’t built for it.”
Levy’s pitch: the AI race is a security-infrastructure race, and AWS’s edge is the stack – classified cloud capacity, public-sector credits, Bedrock (AWS’s managed generative AI platform), embedded engineers and existing infrastructure, bundled together. That bundle buys speed. It also buys AWS gravity, leaving security teams asking where modernization ends and dependency begins – starting with the most sensitive workloads AWS is chasing: classified government and defense data.
AWS Secret Cloud for Industry
AWS Secret Cloud for Industry lets cleared defense contractors, research institutions and other National Industrial Security Program organizations run classified workloads directly on AWS, the company said.
Levy said cleared defense-industrial partners can now reach the same AWS secret cloud regions the Department of Defense uses, with full classified capabilities and familiar services.
Northrop Grumman is the first contractor on the service, and AWS is offering up to $20 million in credits for qualified defense companies. It holds a Defense Information Systems Agency authorization at Impact Level 6, the Secret-classification standard, and cuts provisioning time from months to days, Nextgov/FCW reported.
AWS pitches it as an isolated environment for AI, machine learning and high-performance computing, not just storage, extending its classified-cloud push into a defense industrial base where such work has stayed on-premises.
AWS is also offering up to $1 billion in outcome-based cloud credits through October 2030 for intelligence-community agencies migrating to AWS under its Intelligence Community Accelerated Modernization Framework.
“When you migrate, we share the cost. When you modernize, we cover the engineering,” Levy told the crowd. “We invest alongside you until your most critical workloads are in production.”
The credits cut migration costs. They also pull more intelligence-community workloads into AWS’s model, leaving security teams to weigh portability, audit logging, data governance, key management and customer-control terms.
AWS made a similar pitch with its $1 billion plan to embed thousands of AI engineers with customers building production agentic AI systems. Levy said “95% of enterprise AI fails to reach production,” though AWS cited no source.
“These aren’t advisors — they don’t just hand you a deck and leave,” he said. “They are engineers from our frontier AI teams, including the ones behind the services you use daily.”
Reuters reported AWS plans to send five- or six-engineer pods to customers for 45-day stints, noting Amazon is late to a market Palantir, Salesforce, Anthropic and Google Cloud already serve. Microsoft is chasing the same ground, launching a $2.5 billion Frontier unit July 2 to help customers integrate AI tools, including rivals’ models, with their data, Reuters reported.
AWS’s bet is the bundle: cloud footprint, classified infrastructure, Bedrock and the services those AI systems run on. The security question is how customer access, embedded engineering work, prompts, tool calls and AI actions get approved, logged and audited.
The under-a-minute mitigation demo
In a staged demo, AI agents on Amazon Bedrock detected, analyzed and contained an intrusion in 45 seconds, with “agent actions logged” and decisions “reconstructable” – a demo figure, not an independent benchmark.
AWS also pitched its Kiro coding tool as a compliance-evidence generator. “Every spec Kiro generates, every design decision it documents, every change it records. That is an audit trail,” said Kate Zimmerman, the AWS executive who presented the tools.
The NSA, CISA, FBI and allied agencies warned in 2025 joint guidance about data provenance, trusted revisions and AI lifecycle risks. The Cloud Security Alliance warns AI agents often operate in an identity gray area, borrowing human or shared service-account credentials – and in a CSA survey, 68% of organizations said they couldn’t distinguish AI-agent actions from human ones, while 79% said agents create hard-to-monitor access paths.
CIA Director John Ratcliffe called it “not misplaced” to compare frontier AI capabilities to “digital nuclear weapons.”
“We simply can’t afford to wait for a risk-free approach when it comes to emerging technologies,” Ratcliffe said. “It doesn’t exist. We have to move fast.”
Identity
AWS’s identity and healthcare deployments raise the same governance issues. CLEAR, known for airport security lanes, is integrating CLEAR1 into Amazon Connect, swapping knowledge-based security questions for opt-in verification before callers reach an agent.
Sharecare said AskMD uses Amazon Bedrock to orchestrate multiple foundation models, pulling in FHIR health records and live insurance data.
UK government CTO Sonia Patel told the summit her office stays “cloud agnostic by policy” and uses multiple providers “by design” — though, she added, that “does not mean every platform delivers equally against every requirement.”
AWS remains the largest cloud provider — Synergy Research put its global share at 28% in Q1 2026, ahead of Microsoft’s 21% and Google’s 14%, though both rivals are growing faster.
The remaining questions: how agent identities are managed, how sensitive data is protected, how embedded engineers are constrained, how AI actions are logged, and who’s accountable when cloud, identity and agents converge.
Image Credit: Courtesy of Amazon’s AWS
