AWS Seeks to Remove the Friction Around Government AI Adoption
At AWS Summit DC 2026, Amazon Web Services unveiled new AI modernization and security automation tools built for federal agencies with classified cloud infrastructure and billions of dollars in credits, with thousands of engineers embedded on top. The pitch is speed.
Money Is Up. Skills Are Not.
Federal IT leaders are under pressure to move. Federal News Network data cited by AWS puts AI adoption at 91% of agencies testing or using it, while 72% of IT leaders say their own workforce lacks the skills to use it. The civilian IT budget request for fiscal 2027 sits at $75.7 billion, up from $67.9 billion the year before. Agencies have money to spend and a skills gap that keeps them from spending it well.
AWS Puts Infrastructure and Money Behind New Programs
AWS Secret Cloud for Industry (ASCI) lets defense contractors run contractor-owned classified workloads on AWS infrastructure inside their own isolated environment, a first for the company. It grew out of a multiyear collaboration with the Defense Information Systems Agency and the Defense Counterintelligence and Security Agency. AWS is accelerating this new offering with $20 million in credits over three years; Northrop Grumman is the first partner.
The IC Accelerated Modernization Framework (ICAMF) offers up to $1 billion in cloud credits to help US Intelligence Community agencies migrate workloads to AWS through October 2030. AWS says it will share migration costs and cover engineering work until an agency's most critical workloads are in production.
A separate program extends the same logic to civilian agencies. The GSA OneGov agreement offers up to $1 billion in savings on cloud services, modernization credits, and migration support, a track distinct from the IC-specific ICAMF credits described above.
AWS Forward Deployed Engineering embeds AWS engineers with customer teams to co-build AI systems, backed by a $1 billion investment. AWS says the model compresses development timelines from months to days.
AWS Continuum, announced June 17, automates vulnerability discovery, prioritization, validation, and remediation. It is in gated preview and works alongside existing tools such as Amazon GuardDuty and AWS Security Hub.
AWS Transform targets legacy code and system modernization, a different job than Continuum, which handles security. AWS markets both under the same modernization story, but the mechanics do not overlap.
AWS AI Factory provides dedicated, on-premises AI computing and private AWS environments for training frontier AI models and running inference. It removes the work of building an AI data center from scratch: the agency supplies space and power, and AWS installs and manages the hardware, storage, and networking. The program includes direct access to Amazon SageMaker and leading foundation models, with controls to keep data resident in the agency's chosen location.
AWS Skips the Diagrams and Shows the Receipts
AWS backed the pitch with outcomes, not just architecture diagrams. A federal civilian agency cut document processing from 16 weeks to as little as one day using a multi-agent system on Bedrock in AWS GovCloud, automating 80% to 90% of manual work. Denver's 911 system cut transferred calls by 39% after automating non-emergency call routing.
The setup was a dramatized 3:01am breach at a fictional federal civilian agency. An attacker called Shadow Vertex exploited a public portal and started reconnaissance against an S3 bucket holding millions of citizen records – the kind of incident that in a real, unstaffed overnight security operations center typically runs 47 minutes before a human notices.
In the demo, four AI agents built on Amazon Bedrock AgentCore picked it up instead, each with one job: detection, analysis, containment, and compliance. The detection agent flagged the traffic in seconds. The analysis agent traced six hours of logs, matched the pattern to a MITRE ATT&CK technique, and scored the blast radius at one control, one data bucket, and 94% confidence. The containment agent proposed disabling the credentials and denying the rule, then waited for a single human-approved click before acting.
AWS Continuum took the same incident further. It ran a targeted penetration test that reproduced the SQL injection as evidence, then filed a pull request with the code fix for a human to review. Exploit to containment: 45 seconds, gated by one human approval.
AWS Transform has saved more than 1.6 million hours of manual migration work and rewritten more than one billion lines of mainframe code across its customer base, AWS says.
The Defense Counterintelligence and Security Agency is using it to leave legacy infrastructure for AWS GovCloud, a move AWS projects will save $114 million over three years.
Idemia, whose identity platform runs Department of Motor Vehicles systems in more than 45 states, used Transform to jump from .NET 3.5 to .NET 8 at four times the speed of a manual rewrite, breaking a monolith into microservices on EKS and moving a 5-terabyte database to Aurora PostgreSQL, cutting total cost of ownership by 30% and recovery time from hours to minutes.
CIA Is Making It Easier to Do Business With
CIA Director John Ratcliffe made a version of AWS's own case, from the customer side. He said the agency's technology acquisition timeline used to run three years, including a nine-month security review, and that a new procurement framework and a new procurement executive now target six months, with almost 400 acquisitions closed in the six months since. He credited the change to stripping away red tape and delegating decisions to the lowest possible level.
Ratcliffe also described a broader reorganization. CIA folded its former Directorate of Digital Innovation into a new Directorate of Mission Systems, narrowing its remit to cybersecurity, data, and infrastructure, and elevated its Center for Cyber Intelligence into a full mission center. He set up an Office of Corporate Partnerships as a single point of contact for industry, a direct response, he said, to CIA's reputation as a difficult agency to sell to. He has described frontier AI models to other national security advisors as comparable to digital nuclear weapons and said CIA's approach is to take smart risk and correct course rather than wait for a risk-free option that does not exist.
The UK Pairs AWS Results With a Multi-Vendor Policy
UK Government Chief Technology Officer Sonia Patel offered a counterpoint from the same stage. The UK is agnostic by policy, she said: it uses multiple cloud providers by design, not because any single vendor locked it in. Her three non-negotiables, regardless of provider: security and compliance, AI as a first-class capability, and the ability to operate at public sector scale.
The UK's AWS-run results are real. gov.uk Chat improved from roughly 76% accuracy in early pilots to 90% in production over 18 months. One Login now serves 14 million people across 120 government services. England's National Lung Cancer Screening Program has screened more than one million people and caught more than 3,000 cancers early enough to treat. HM Revenue and Customs is spending £473 million to leave three legacy data centers for AWS. Patel also confirmed a hands-on legacy modernization session using AWS Transform, with participating agencies reporting work that used to take days finishing in minutes.
Genesis Mission Extends the Same Trade to Science and Defense
AWS is underwriting a parallel push into federally funded research. The Genesis Mission, launched by a presidential executive order last November, aims to connect 17 national laboratories and roughly 700 consortium partners into a single AI-powered research platform. AWS committed up to $100 million in credits split across two programs: the AWS Genesis Accelerator (up to $50 million for Department of Energy labs and their private sector partners) and the AWS Warfighter Capability (up to $50 million for the Department of War and the defense industrial base). AWS also announced a $50 billion investment to expand AI and supercomputing infrastructure across GovCloud, Secret, and Top Secret regions, targeting 1.3 gigawatts of AI capacity.
Our Take
Public sector IT has two problems that predate any AI pitch: enormous and poorly understood application sprawl that goes part and parcel with this and is endemic in public sector eating the budget and not enough staff who can run what replaces it. Fixing both is the mandate, regardless of which vendor helps.
AWS's programs, plus Genesis Mission and GSA OneGov, are built to solve exactly that pair of problems, and they lower the switching cost enough that choosing AWS starts to look like the fast path. That is a reason to look closely at what Google Cloud, Microsoft Azure, and Oracle are offering federal and public sector agencies on the same two problems, not a reason to skip the comparison.
Forward Deployed Engineering deserves a harder look than the others. Embedding AWS engineers inside an agency's build touches data handling, security clearance, and procurement rules that vary by agency. Weigh the program against the agency's own regulatory and compliance structure before signing, not just against the speed it promises. Forward Deployed Engineering can increase vendor lock-in and client dependency on Amazon.
AWS's Nitro architecture, hardware-level isolation built into every instance, and the ability to choose exact geographic regions where data is stored, processed, and maintained are real advantages for agencies handling classified or sensitive workloads on AWS and serious about data sovereignty. CIOs who go the AWS route should treat these as inputs while still considering what hardware-level protections Azure Government, Google Distributed Cloud, and Oracle offer before treating them as decisive.
CIOs should move quickly on two fronts that are independent of vendor choice: retiring legacy technology and building AI skills in house. Both are foundational to closing the gap between AI ambition and workforce readiness described above, and progress on either pays off no matter which cloud provider is ultimately selected.
CIOs should move more deliberately on the vendor decision. Before committing to AWS, validate its claimed savings and timelines against at least one competing offer from Google Cloud, Microsoft Azure, or Oracle, and confirm that programs such as Forward Deployed Engineering, data residency controls, and hardware-level isolation align with the agency's own regulatory and compliance requirements. Close the vendor decision only after that comparison is complete.
Want to Know More?
Amazon Bets Big on Agentic AI Development With $50 Billion OpenAI Investment
Amazon's Bedrock: Unlocking the Gateway to Powerful Open LLMs