AI Week in Review - 4/19/25

Public Sector

Author

Chris Kraft
April 19, 2025

Good morning, AI enthusiasts. While other AI labs have been rushing out model after model, Anthropic has been eerily quiet since the launch of 3.5 Sonnet in June of 2024.

But the startup’s long silence is reportedly about to be broken — with a new hybrid reasoning system launching in the ‘coming weeks’ that could be right on time for the industry’s shifting model meta.

In this week’s edition:

  • Agentic AI in the Public Sector

  • U.S. House Investigates DeepSeek

  • The latest public (and private) sector news

The Highlights

Source: ChatGPT

The Story
The AI landscape is quickly moving more towards agentic AI. These solutions leverage reasoning, memory, and tools to perform autonomous, goal-driven tasks. The public sector needs to be prepared for these new agentic solutions.


More Details
But what does this really mean - what is agentic AI and why should you care?

  • What are agents? Agents are able to decomposes problems, uses tools, asks clarifying questions, adapt in real-time and evolve the output to meet mission objectives.

  • What are the building blocks of agentic solutions? In this video, Andrew Ng outlines four agentic AI design patterns: Reflection, Tool Usage, Planning, and Multi-Agent Collaboration.

  • Are there any unique considerations with agentic AI? Some important considerations include memory, evaluation, and human-in-the-loop. These are key to building agentic solutions and moving them to production.

What’s Next?
This article provides more details on how we start preparing to leverage agents in the public sector. If you’re looking for training and other resources, this Google Doc includes over 70 free agentic AI resources.

The Story
The U.S. House Select Committee on the Chinese Communist Party just released their report on DeepSeek. The report starts off noting that the U.S. lead over China is not 18-months as claimed by some, but something closer to 3-months.

More Details
The report included the following findings:

  • DeepSeek funnels Americans’ data to the PRC through backend infrastructure connected to a U.S. government-designated Chinese military company.

  • DeepSeek covertly manipulates the results it presents to align with CCP propaganda, as required by Chinese law.

  • It is highly likely that DeepSeek used unlawful model distillation techniques to create its model, stealing from leading U.S. AI models.

  • DeepSeek’s AI model appears to be powered by advanced chips provided by American semiconductor giant Nvidia and reportedly utilizes tens of thousands of chips that are currently restricted from export to the PRC.

The Details

Federal

What America gets wrong about the AI race
To maintain its AI edge, the U.S. must shift focus from model performance to widespread adoption across government, military, and global markets—backed by infrastructure, regulation, and strategic partnerships.

DeepSeek unmasked by House Committee
A congressional report warns that DeepSeek poses national security threats by funneling U.S. data to China, spreading propaganda, and allegedly using stolen U.S. tech and restricted chips to build its model.

Nearly 50 House Democrats say no to unauthorized AI in Trump administration
House Democrats have demanded the Trump administration halt unapproved AI use—particularly DOGE’s deployment of Musk’s xAI tools—citing legal, ethical, and security concerns over employee monitoring and data access.

U.S. tries to crush China’s AI ambitions with chips crackdown
The Trump administration has imposed new restrictions on Nvidia and AMD AI chip exports to China, aiming to stifle Beijing’s AI progress and reinforce U.S. tech dominance amid growing geopolitical tensions.

Joint Staff pursues ‘major step forward’ to enhance ORION force management platform with AI
The Joint Staff is partnering with BigBear.ai to integrate advanced AI into its ORION platform, aiming to streamline global force management and improve real-time military decision-making across the Pentagon.

AI may aid screening for opioid use disorder
An NIH-backed study showed that an AI screening tool embedded in hospital workflows effectively identified patients at risk for opioid use disorder and significantly reduced hospital readmissions, cutting healthcare costs.

Claude goes to Washington: New alliance accelerates AI access across federal agencies
Palantir, Anthropic, and Google are partnering to deploy Claude across federal agencies by leveraging FedRAMP-compliant infrastructure, accelerating secure government access to advanced generative AI tools.

SSA is rolling out a new chatbot for employees
To offset deep staffing cuts, the Social Security Administration is launching an internal generative AI chatbot to assist employees with tasks like research, content creation, and coding—though concerns about oversight and testing remain.

OSTP taps policy scholar Dean Ball as AI, emerging tech advisor
Dean Ball, a policy scholar with ties to conservative institutions and a focus on executive power, has joined OSTP as a senior AI advisor in the Trump administration.

NGA goes all-In on AI: 2025 marks major push to transform intel with AI
The National Geospatial-Intelligence Agency is declaring 2025 the “year of AI,” ramping up adoption of multimodal and generative AI for missions like targeting, warning, and intelligence fusion through new leadership and agile procurement strategies.

Former Army AI leader tapped as Pentagon’s next CDAO
Douglas Matty, a seasoned Army AI leader, has been appointed as the Pentagon’s Chief Digital and AI Officer, signaling a renewed push to accelerate AI and data integration across defense operations.

State / Local

PA lawmakers map out AI future from Pittsburgh’s tech corridor
At a Pittsburgh hearing, Pennsylvania lawmakers and experts explored how to position the state as a national AI leader by leveraging local talent, infrastructure, and energy innovation amid rising AI adoption.

What Is Agentic AI? A Look at New Tech Reshaping Government
Agentic AI, which enables autonomous, goal-driven decisions without human prompts, is beginning to reshape government operations, prompting leaders to balance innovation with human oversight and policy safeguards.

New Jersey taps the private sector to facilitate AI innovation and talent
New Jersey is launching two major programs—a $3.8M innovation grant and a $500M tax credit—to drive AI development, workforce growth, and public-private collaboration across the state.

Utah enacts mental health chatbot law
Utah’s new HB 452 law imposes strict rules on AI mental health chatbots, requiring transparency, user protections, and compliance frameworks to safeguard Utah users and ensure ethical AI use in mental health.

Kentucky city wraps AI project collecting community planning feedback
Bowling Green, Kentucky used AI to analyze over a million pieces of public input for its 25-year community plan, revealing strong consensus on key priorities like healthcare access and historic preservation.

Pennsylvania county looks to fill AI advisory council
Montgomery County, PA, is forming an AI advisory council to guide ethical, transparent use of AI in government operations, aiming to enhance public services while building resident trust.

NJ schools use AI to accelerate French lessons
With state grants and clear guidelines, New Jersey schools like Gateway Regional are using AI to personalize language learning, streamline teaching, and build students’ critical thinking and conversational skills.

Police use AI to generate photo of suspect who tried kidnapping a teen girl
Goodyear, Arizona police used AI to create a highly realistic suspect image based on witness input in a broad-daylight attempted kidnapping, marking a first-time use of the tech to aid their investigation.

Report shows rise in online scams and AI-driven identity theft in Alaska
A new BBB report highlights a growing surge in AI-enhanced online scams and identity theft in Alaska, with nearly 600 incidents reported since 2021 and increasingly convincing tactics used by fraudsters.

AI technology at AZ assisted living facilities help protect residents
AI-powered radar devices in Arizona assisted living centers have reduced resident falls by up to 70%, alerting staff to unusual movements without compromising privacy.

California crosswalks hacked with AI voices of Zuckerberg and Musk
AI-generated deepfakes of Elon Musk and Mark Zuckerberg were broadcast from hacked Bluetooth-enabled crosswalks in Palo Alto and Redwood City, prompting cities to disable voice features amid repairs.

Illinois bars AI in workers comp proceedings
Illinois has banned the use of AI tools and recording devices in workers' compensation proceedings, warning participants they will be removed if found using them during hearings.

AI helps prepare the first voice you hear after calling 911 in Johnson County
The Johnson County Sheriff’s Office is using CommsCoach AI to simulate 911 calls and accelerate dispatcher training and call review, aiming to improve emergency response readiness.

International

NATO inks deal with Palantir for Maven AI system
NATO awarded Palantir a fast-tracked contract for the Maven Smart System, aiming to boost battlefield decision-making and intelligence through AI-driven tools like language models and machine learning.

Thailand rolls out AI Robocop to safeguard crowds
Thailand debuted its first AI police robot, “Cyborg 1.0,” at the Songkran festival, using facial recognition, behavior monitoring, and weapon detection to boost public safety through real-time surveillance.

Russia seeds chatbots with lies
Russia is exploiting AI vulnerabilities to inject propaganda into chatbots, exposing how easily bad actors can manipulate generative models and influence public opinion through “LLM grooming.”

AI and national security
RUSI’s report outlines how AI is reshaping UK national security—from tactical warfare to strategic planning—urging investment, ethical oversight, and preparation for a rapidly approaching AGI future.

Hawkish AI? Uncovering DeepSeek’s foreign policy biases
CSIS researchers found that DeepSeek-V3 displays a statistically significant bias toward recommending escalatory foreign policy actions—especially against Western democracies—highlighting the national security risks of using unvetted LLMs in crisis scenarios.

Everything Else

OpenAI releases 4.1 models
OpenAI has released GPT-4.1, mini, and nano—three upgraded models offering stronger coding skills, improved instruction following, and support for up to 1 million tokens for long-context tasks.

Is OpenAI building a social network?
OpenAI is developing a prototype social network built around ChatGPT’s image generation, signaling a bold move into social media and intensifying competition with X and Meta.

IAPP releases AI Governance Profession Report
AI governance is becoming a strategic imperative in 2025, as enterprises face mounting regulatory pressures and recognize responsible AI as essential for competitiveness and long-term scalability.

OpenAI releases a practical guide to building agents
This guide from OpenAI offers actionable frameworks and best practices to design AI agents that handle complex workflows autonomously, making them ideal for scenarios where traditional automation falls short.

J&J pivots its AI strategy
Johnson & Johnson is narrowing its generative AI focus to the highest-impact use cases—like drug discovery and supply chains—after finding only a fraction of early pilots delivered real value.

10 most used GenAI tools in the enterprise
From ChatGPT to Claude, enterprises are embracing a wide range of generative AI tools—with chatbots, copilots, and visual generators leading the way in productivity, creativity, and search applications.

AI Agent Governance: A Field Guide
As autonomous AI agents advance rapidly, experts warn of profound societal and security risks—and call for urgent governance to align, control, and safely integrate these emerging digital actors.

Smarter workdays: Claude’s new research and Workspace integration boosts collaboration
Anthropic’s Claude now offers agentic web research and Google Workspace integration, enabling faster, context-aware insights across documents, email, and calendar—streamlining personal and professional decision-making.

How AI is using facial recognition to help bring lost pets home
AI-powered facial recognition technology from Petco Love’s "Love Lost" database has helped reunite 100,000 lost pets with their families by matching photos using unique animal features.