Weekly Roundup - May 4-8, 2026

ARTIFICIAL INTELLIGENCE — ADOPTION, GOVERNANCE & SECURITY

Survey: More Than Half of Federal Agencies Now Planning Agentic AI Pilots. A Market Connections survey of more than 200 technology executives across civilian and defense agencies, sponsored by ServiceNow and released May 5, finds that 53 percent of agencies are exploring or actively planning pilots of agentic AI, with another 15 percent already implementing or having fully completed agentic AI systems. Yet even as agencies race toward deployment, the survey surfaces a critical governance gap: respondents consistently identified inadequate oversight policies and a significant disparity between the perceived importance of governance frameworks and their actual implementation. Data readiness emerged as the most durable practical barrier and agencies struggle to move pilots from sandbox to production when underlying data architectures are not designed to support AI at scale. The findings capture a federal AI environment defined by momentum at the top and structural fragility underneath.

White House Weighs Reining In Contractors' Control Over How Agencies Use AI. The federal government is circulating draft policy documents containing language that would formally assert the government's authority to determine how privately developed AI technology is used in federal operations without contractual restrictions imposed by vendors. Two sources familiar with their development told Nextgov/FCW that the draft language centers on the principle that it is for the democratically elected government to determine what is a lawful and appropriate use of a particular technology, not solely a company. The drafts also examine managing cybersecurity threats posed by frontier AI models. Whether the language will crystallize into an executive order remains unclear, but the direction is unambiguous: the administration intends to assert greater governmental sovereignty over AI procurement and use, a posture directly connected to the ongoing Anthropic-Pentagon dispute.

Agency Leader Says AI Is Helping Resource-Strained Workforce Identify More Fraud. Speaking at the UiPath Fusion Public Sector conference on May 5, Jeneen Iwugo, acting director for the Centers for Medicare and Medicaid Services Center for Program Integrity, said AI is dramatically expanding her workforce's capacity to detect healthcare fraud. With an annual budget of approximately $1 billion to address an estimated $100 billion to $300 billion fraud problem, and roughly 500 staff reviewing four to five million claims daily, Iwugo said AI tools enable CPI to work at a scale impossible through human review alone. CPI's FY2024 return on investment was $14.6 for every $1 invested, generating an estimated $26.3 billion in savings — an improvement over the prior year. Iwugo noted that the Trump administration's anti-fraud priorities have created a favorable environment for AI adoption, framing the technology as a point of alignment between mission need and current administration policy.

Operational Technology Providers Are Feeling 'Annoyance' at Exclusion from Anthropic's Mythos Rollout.  Anthropic's limited release of its Mythos Preview model is generating frustration among operational technology providers excluded from the select group given early access. The model has drawn intense attention from government officials and the national security community due to its autonomous exploit development potential. The Pentagon had previously labeled Anthropic a supply chain risk, leading to a White House-ordered governmentwide phaseout of Anthropic technology, which is a ban subsequently challenged in court and temporarily enjoined. The standoff between Anthropic's use restrictions and the government's desire for unfettered AI capability is reshaping how the administration thinks about frontier AI governance, sovereign technology access, and the limits of commercial vendor authority over national security applications.

White House Weighs Safety Reviews for Frontier AI Models. National Economic Council Director Kevin Hassett confirmed on May 6 that the White House is studying a potential executive order that would establish a safety review process for frontier AI models before public release and likening the proposed framework to FDA drug approval. His comments follow Anthropic's limited release of its Claude Mythos Preview model, restricted due to the model's ability to identify and potentially exploit software vulnerabilities. Hassett said National Cyber Director Sean Cairncross is coordinating with Anthropic to ensure Mythos undergoes rigorous testing before broader release. The remarks represent a notable pivot: the Trump administration, which had aggressively deregulated AI and banned Anthropic products from federal use, is now signaling a more interventionist posture on frontier model safety — driven by recognition that the most capable models carry genuine national security risk.

CISA Offers Guide for 'Careful' Agentic AI Adoption. CISA and a coalition of international cybersecurity partners released formal guidance on May 7 titled Careful Adoption of Agentic Artificial Intelligence Services — the first dedicated federal guidance document specifically targeting the cybersecurity risks of autonomous AI agents. The guide identifies four primary risk categories: expanded attack surfaces, privilege creep from overly broad permissions, behavioral misalignment in AI decision-making, and limited visibility from incomplete event records. Core recommendations urge organizations to deploy agentic AI beginning in low-risk environments, avoid granting agents broad access to sensitive data or mission-critical infrastructure, and integrate agentic AI into existing cybersecurity and risk management frameworks rather than treating it as a standalone capability. Acting Director Nick Andersen said CISA is committed to ensuring AI adoption aligns with President Trump's Cyber Strategy for America.

DOD CDAO Wingman to Scale Automation Across Military. The Department of Defense's Chief Digital and Artificial Intelligence Office is expanding its CDAO Wingman task automation platform department-wide, following deployment within the Army's procurement office where it automated more than 150 workflows, saved over 687,000 work hours, and avoided more than $37 million in costs. Speaking at the UiPath Fusion conference on May 5, CDAO War Data Platform Director Elizabeth Chirico said Wingman combines AI, large language models, generative AI, robotic process automation, and low-code tools to let personnel build custom automations tailored to specific mission needs. The CDAO received authority to operate in February 2026 to deploy the suite department-wide and is extending access across all military departments. A new position description-writing bot is nearing release on the platform.

AGENCY MODERNIZATION & INNOVATION

OPM Looks to Modernize Retirement Services With AI. OPM Director Scott Kupor told the UiPath Fusion conference on May 5 that artificial intelligence is the most direct path to modernizing the agency's deeply paper-intensive retirement services operation, which still relies on mailed documents between HR departments, payroll offices, and OPM. He described AI-powered call center automation and digital self-service tools as enabling employees to shift from routine inquiries to complex, judgment-intensive cases. Kupor also highlighted the newly launched USA Class tool, which uses AI trained on thousands of existing position descriptions to automate federal job description creation compliant with OPM classification standards — available to all agencies through USA Staffing at no additional cost. He characterized these as iteration one of a broader modernization agenda he intends to accelerate across the entire federal human capital lifecycle.

Stanton Outlines GSA Vision for Procurement Automation Ecosystem. Acting GSA FAS Commissioner Laura Stanton described to the 2026 ITVMO Annual Summit on May 6 a connected procurement automation ecosystem designed to reduce manual workload, accelerate buying, and increase compliance confidence across the acquisition lifecycle. Stanton defined the vision as a connected operating environment that brings people, process, data, technology, and governance into a system that operates as one — giving GSA visibility not only into acquisition competitiveness but into how well contracts are actually delivering on mission. She identified market research, clause selection, responsibility checks, inter-system data movement, and delivery tracking as prime automation candidates. The ecosystem builds on a prior GSA request for information on an AI-driven acquisition platform and aligns with GSA's restructuring of FAS into five portfolios, with a new office dedicated to automation and AI.

Commerce AI Center Will Evaluate Google DeepMind, Microsoft, and xAI Models. The Commerce Department's Center for Artificial Intelligence Standards and Innovation has renegotiated voluntary agreements with Google DeepMind, Microsoft, and xAI to allow CAISI to evaluate their frontier AI models in classified environments — a significant expansion of the federal government's ability to assess the national security risks of the most powerful commercial AI systems. The agreements align with the Trump administration's AI Action Plan and build on prior voluntary commitments. CAISI will study models with reduced or removed safeguards to understand their unmitigated capabilities. Director Chris Fall called the initiative essential to understanding frontier AI and its national security implications. The agreements follow CAISI's prior evaluation of Chinese model DeepSeek, which underperformed U.S. models in accuracy, security, and cost efficiency.

Pentagon Will 'Never Again' Rely on a Single AI Provider, Official Says. Defense Under Secretary for Research and Engineering Emil Michael declared at the SCSP AI+ Expo on May 7 that the Pentagon will never again be single-threaded with any one AI model — a direct response to the strategic vulnerability exposed by the Anthropic-Pentagon conflict. Michael described recent AI cooperation agreements with eight leading AI developers as a concrete step toward a diversified technology stack. He said models deployed on defense networks must be tuned for national security purposes, and that any vendor guardrails must be consistent with the government's values, mandate, and restrictions. The remarks frame DoD's AI procurement approach as a portfolio strategy that prioritizes flexibility, sovereign control, and the ability to shift among providers based on mission requirements.

CYBERSECURITY & GOVERNANCE

CISA Unveils CI Fortify to Help Secure Critical Infrastructure During Conflicts. CISA released its CI Fortify initiative this week, providing structured guidance to critical infrastructure operators on maintaining and recovering operations during a major conflict involving adversarial cyber attacks on civilian systems. The guidance is explicitly oriented toward Chinese infiltration of non-military critical infrastructure, which U.S. officials assess is ongoing and intended to enable sabotage in a Taiwan conflict scenario. CI Fortify directs operators to assume that telecommunications, internet, vendor relationships, and upstream dependencies will be unreliable in a conflict environment, and that threat actors will have some access to operational technology networks. Acting Director Andersen encouraged organizations to collaborate with CISA on strengthening defenses against both opportunistic and state-sponsored threat actors.

IBM Security Executive Emerges as Possible Contender to Lead CISA. Tom Parker, a cybersecurity executive at IBM, has emerged as a potential candidate to lead CISA following Sean Plankey's nomination withdrawal last month, according to a person familiar with the matter. Parker does not have prior government experience — a characteristic sources suggest may align with the Trump administration's current preferences for agency leadership. CISA has been without a Senate-confirmed director for the entirety of Trump's second term, has lost roughly a third of its workforce, and faces a proposed $707 million budget cut for FY27. Acting Director Nick Andersen has stabilized operations and secured approval for 329 mission-critical new hires, but the governance gap created by the absence of confirmed leadership continues to affect CISA's institutional standing at a moment of elevated cybersecurity threat.

LEADERSHIP, MANAGEMENT & WORKFORCE

White House Taps Thomas Flagg as Deputy Federal CIO OMB has appointed Thomas Flagg, currently the Education Department's CIO, to serve as deputy federal CIO, reporting to Federal CIO Greg Barbaccia. Flagg takes over from Jay Teitelbaum, who had been serving in an acting capacity. He brings more than 11 years of experience at the Department of Labor and a subsequent tenure as Education CIO since October 2024. As deputy federal CIO, Flagg will advise OMB senior leadership on federal IT policy, oversee modernization and agency engagement, and shape data governance policy across the executive branch. The appointment fills a sustained leadership gap at a moment when governmentwide AI adoption, FedRAMP reform, and technology consolidation are all in active flux.

Why Effective Leaders Get Branded as Problems. When a leader creates friction, organizations default to a single explanation: the leader needs to change. In reality, that friction usually comes from one or a combination of four different sources—capability, perception, identity, or system. Because those look similar on the surface, organizations tend to categorize them under one bucket, behavior and make high-stakes decisions based on that assumption. The cost is not just ineffective development. It is flawed promotion decisions, stalled succession pipelines, and the quiet loss of high-impact leaders who are labeled as “difficult” when they are, in fact, misread.

Now Is the Time for Courage. In the face of political, economic, and technological uncertainty, business leaders are often reluctant to take bold action. Some freeze, too overwhelmed to make decisions. Many hunker down, hoping to wait out the chaos. Others retrench, trying to protect their organization’s future and their careers. However, research shows that the old adage is true: Fortune favors the brave, not the cautious.

Communicating with Confidence When You’re Under Pressure. Communicating clearly sets you up to have the leadership impact and influence you need to drive change. But what if you’re running on empty? Expressing your ideas and giving direction when you’re sleep-deprived, burned out, or simply overwhelmed can feel nearly impossible. So, what helps? Leadership development coach Muriel Wilkins, author of Leadership Unblocked and host of the podcast Coaching Real Leaders, talks us through communication techniques that meet you where you’re at mentally and emotionally so that you can rise to the moment (even when you’re worried you can’t).

The Best Leaders Embrace the Role of Supporting Character. In 2024, Y Combinator’s Paul Graham wrote that Silicon Valley leaders must directly impose their vision on employees—a style of leadership he called “founder mode.” His essay attracted 20 million views on social media and launched dozens of think pieces (including at HBR).

THIS WEEK @ THE CENTER 

RECENT BLOGS                                             

• The Mission Generation: Rethinking Talent, Purpose, and Public Service by Michael J. Keegan. I host The Business of Government Hour to highlight ideas that challenge how we think about leadership, talent, and service. My conversation with Gupta did exactly that. The Mission Generation is not a theory of work. It is a reframing of how people make decisions about their careers, and how institutions must respond. At its core, the conversation addressed a simple tension. People want purpose. Systems still reward something else.
• How Can Responsible Application of AI Help Child Welfare Agencies Serve Children and Families More Effectively? by Dan Chenok.  A new report explores how artificial intelligence can help child welfare workers across the country to better support their constituents — not by replacing human judgment, but by freeing up people to spend more time on delivering needed services and less time on repetitive processes.
• From Principles to Practice: Governance Actions That Amplify Human Agency by Dr. David Bray. What specific actions can leaders at local, state, and national levels take to ensure that technology serves as a powerful amplifier of human capability?

ICYMI – This week Michael J. Keegan welcomed Arun Gupta to discuss his new book, The Mission Generation and explore a challenge that sits at the center of public service today. Talented people want to serve. The systems we have built make that choice unnecessarily hard. He explains the Compass Capital framework for rethinking how we measure career success. He describes four resistances that block people from pursuing meaningful work. He offers the Mission Flywheel model for turning purpose into action. One line from Gupta that I keep returning to: "Government is selling careers. Young people are buying experiences."

WEEK IN REVIEW

This was the week agentic AI shifted from a technology priority to a governance problem in the federal government. A Market Connections survey released May 5 found more than half of federal agencies now planning or actively piloting agentic AI systems. Yet respondents were equally clear about what is missing: oversight policies lag deployment, and data architectures are not ready for the scale agencies are moving toward. CISA's release on May 7 of the first federal guidance dedicated to agentic AI security acknowledged the same reality in official terms. Agencies are fielding autonomous systems before they have built the frameworks to oversee them.

The Anthropic Mythos Preview model drove a separate and significant policy story. The White House indicated it may pursue an executive order establishing safety reviews for frontier AI models prior to public release. At the same time, the administration circulated draft policy language designed to confirm the government's authority to use acquired AI technology on its own terms, without vendor-imposed restrictions on how those tools may be applied to federal missions. The Pentagon's under secretary for research and engineering added his voice to the same theme, stating plainly that the department will not again depend on a single AI provider for its operations.

The UiPath Fusion Public Sector conference on May 5 offered the week's most concrete look at AI in practice across agencies. OPM Director Kupor described plans to automate retirement services processing and federal job description writing. GSA Acting FAS Commissioner Stanton laid out a procurement automation framework intended to connect currently fragmented acquisition systems. The DoD's CDAO reported that its Wingman automation platform has saved more than 687,000 work hours and avoided over $37 million in costs within the Army's procurement office alone.