CISA Layoffs, AI Chatbots in Government, and Utah's Age Verification Law: Tech Policy Updates

[00:00:01] It's Wednesday, March 12th, 2025, and I'm Dave Sobel. Three things to know today. Cybersecurity jobs cut, AI hired, and kids get ID'd. Welcome to the future of tech policy. ChatGPT isn't killing Google search, and AI lies more than you'd think. And Microsoft and OpenAI, a rocky relationship while AI prices tumble. This is the Business of Tech.

[00:00:28] Former Trump administration cybersecurity official Sean Plancky has been nominated to lead the Cybersecurity and Infrastructure Security Agency, also known as CISA. His nomination was announced on March 12th, 2025, and has been forwarded to the Homeland Security and Governmental Affairs Committee for review.

[00:00:46] Plancky previously held significant roles in cybersecurity, including Acting Assistant Secretary at the Department of Energy and Director for Maritime and Pacific Cybersecurity Policy at the National Security Council during the first Trump administration.

[00:01:02] Plancky has laid off over 100 employees from the U.S. Government's cybersecurity agency, CISA, including members of its RED team, which is responsible for simulating cyber attacks to identify security vulnerabilities. The layoffs occurred in late February and early March, and sources indicate that more than 80 employees involved in continuous monitoring and between 30 and 50 incident response staff were affected.

[00:01:28] CISA spokesperson declined to comment on the cuts, but stated that the RED team remains operational. This round of layoffs marks the third known reduction in staff at CISA since January, raising concerns about the agency's ability to maintain cybersecurity amid ongoing federal budget cuts. Doge has been rolling out a custom chatbot called GSA to employees of the General Services Administration aimed at automating various government tasks.

[00:01:58] According to a report from Wired, the chatbot can draft emails, create talking points, summarize text, and write code. Employees have the option to choose from three models, CLAWD Haiku 3.5, CLAWD Sonnet 3.5 version 2, and Meta Llama 3.2. This implementation comes amid significant job cuts in the federal government, with the GSA reportedly laying off over 1,000 workers itself. Internal memos have cautioned employees against inputting sensitive information into the chatbot.

[00:02:27] The initial testing of GSA I took place among a smaller group before it was made available to about 1,500 workers with plans for future expansion. Utah has made history by becoming the first state to pass legislation requiring App Store operators to verify the ages of users and obtain parental consent from minors downloading apps.

[00:02:49] The App Store Accountability Act aims to enhance online safety for children amidst a growing movement across various states advocating for similar regulations. This new law comes as lawmakers have been trying to establish design regulations to protect minors, although many previous efforts have faced legal challenges. Major tech companies, including Meta and Snap, have expressed support for the initiative, emphasizing that parents desire a comprehensive solution for verifying their children's ages.

[00:03:16] However, some organizations warn that such age verification requirements could infringe on privacy rights and chill access to protected speech for all users. The Supreme Court is expected to examine age verification issues this year, particularly concerning adult content websites. Why do we care? This raises concerns about federal cybersecurity posture.

[00:03:38] If CISA's capabilities are weakened, it could create a trickle-down effect, leading to increased threats in the private sector as federal infrastructure and intelligence sharing weakens. From an industry standpoint, the cuts to CISA's workforce, especially in continuous monitoring and red teaming, could lead to reduced proactive security efforts and slower responses to cyber incidents.

[00:03:59] It may push organizations, including critical infrastructure providers, to seek alternative security services, a potential opportunity for managed security service providers to fill the gap. However, it also raises concerns about an increased attack surface and higher risk for supply chain partners. The GSA layoffs underscore a broader trend, government budget constraints leading to tech-driven workforce reductions. This aligns with corporate trends where AI is replacing administration and knowledge-based roles.

[00:04:29] If GSA suffers from data privacy concerns or inaccurate responses, it could slow down AI adoption in government and cause a chilling effect from the private sector. Government typically moves slow for a reason. Utah's new law on age verification for app stores is significant because it represents a growing regulatory push for online safety laws.

[00:04:51] For tech companies, compliance with age verification and parental consent requirements introduces new operational challenges, particularly around user authentication, data privacy, and content moderation. This episode is supported by Flexpoint. Flexpoint offers a purpose-built payment solution from managed service providers, automating billing operations to enhance efficiency and cash flow.

[00:05:14] With features like accounts receivable automation, branded client portals, and secure same-day payments, Flexpoint streamlines financial management. Integrations with accounting software such as QuickBooks and Xero, as well as professional services automation tools like ConnectWise and Autotask, ensure seamless data synchronization. Experience improved cash flow and client satisfaction with Flexpoint's comprehensive platform.

[00:05:39] Learn more at getflexpoint.com slash msp-radio. New data from Google indicates that search queries are continuing to grow, countering concerns that OpenAI's ChatGPT is impacting Google's business. The company reported over 5 trillion searches annually, with Barclays analysts estimating a growth rate exceeding 20% since ChatGPT's launch in late 2022.

[00:06:05] Active users of ChatGPT have surged from 100 million weekly in November 2023 to a reported 400 million in February 2025. Despite these figures, analysts believe that OpenAI is not currently affecting Google's search volume. A new benchmark developed by researchers from the Center for AI Safety and Scale AI aims to measure how much artificial intelligence models lie.

[00:06:31] The Model Alignment Between Statements and Knowledge benchmark, or MASK, evaluates the ability of AI models to knowingly deceive users. In testing 30 frontier models using over 1,500 queries designed to elicit lies, the researchers found that larger models do not necessarily correlate with higher honesty.

[00:06:51] For instance, Grok 2 exhibited the highest dishonesty rate at 63%, while Claude 3.7 Sonnet had the highest honesty rate at 46.9%. The researchers emphasized the risk posed by dishonest AI, including potential legal and financial harm, and highlighted the need for a reliable method to assess honesty in AI systems. The benchmark dataset is now publicly available, aiming to foster advancements toward more truthful AI.

[00:07:21] Why do we care? The data indicating that Google search queries continue to grow despite ChatGPT's rise challenges the narrative that AI-powered chat interfaces are replacing traditional search. For MSPs and IT service providers and their customers, this signals that SEO strategies and online discoverability remain critical for businesses. The MASK benchmark has direct implications for businesses deploying AI tools.

[00:07:47] The findings, that some models exhibit high dishonesty rates, underscores the risk of AI hallucinations, which are particularly concerning for industries reliant on accurate data, such as legal and compliance sectors, where false information could have legal consequences, financial services where incorrect AI-generated insights could lead to financial loss, and cybersecurity and IT services where deceptive AI responses could create security vulnerabilities. These developments reinforce that AI isn't a perfect replacement for existing tools.

[00:08:17] It's an augmentation that requires oversight. Providers should position themselves as advisors on AI governance, security and business intelligence. So let's look at a bunch of platforms. Amazon Web Services has announced the launch of DeepSeq R1 as a fully managed large language model in Amazon Bedrock, marking a significant step in cloud-based AI.

[00:08:40] The model offers advanced reasoning skills and exceptional contextual understanding, making it suitable for complex tasks such as data analysis and coding. Notably, the model is reported to be 90 to 95% more cost-effective than similar offerings in the market. Since its availability in late January, thousands of customers have already started deploying DeepSeq R1 through Amazon Bedrock. AWS emphasizes the importance of integrating safety measures such as Amazon Bedrock guardrails to ensure responsible usage of generative AI.

[00:09:11] A new artificial intelligence system named MANUS developed by MANUS AI has been launched in China and is attracting significant attention. The agent operates independently without human oversight, performing tasks such as analyzing financial transactions and screening job candidates with remarkable speed and precision. Early users have reported mixed experiences, citing issues like error messages and inconsistent performance, despite MANUS's potential to outperform other AI tools.

[00:09:39] Co-founder and chief scientist Yai-chou-Pik-Zhi demonstrated MANUS's capabilities in Shenzhen, showcasing its ability to initiate tasks autonomously, assess new information and adapt its strategies. Microsoft announced an upgrade to the Think Deeper feature of its co-pilot assistant, powered by a more advanced artificial intelligence model from OpenAI.

[00:10:00] This enhancement follows the company's recent decision to eliminate usage limits for Think Deeper, which allows the assistant to spend approximately 30 seconds or more on complex inquiries. The previous model, OpenAI's 01, has been replaced with the 03 Mini High model, noted for its cost efficiency and capability to handle intricate questions. Microsoft AI CEO Mustafa Suleiman highlighted that all users can now access this improved functionality for free.

[00:10:29] Microsoft is also intensifying its efforts to compete with OpenAI by developing its own advanced artificial intelligence models. The tech giant has created AI reasoning models similar to OpenAI's offerings, but tensions have arisen as OpenAI has reportedly denied Microsoft's requests for technical specifics. Additionally, Microsoft has introduced a new family of models named MAI, which aim to rival OpenAI's technology and may be available through an application programming interface later this year.

[00:10:58] The company is also exploring alternative AI models from various firms, including XAI and Meta, to replace OpenAI technology in its co-pilot product. Microsoft has invested approximately $14 billion in OpenAI to date. Anthropic launched an updated platform that enhances collaboration across teams using AI, notably introducing features for its clawed AI assistant.

[00:11:22] The new Anthropic console allows non-technical staff to work alongside developers on AI prompts, addressing challenges in coordinating prompt engineering. This move aims to democratize AI, making it accessible to all employees, not just technical specialists. The platform now includes extended thinking controls for deeper reasoning, balancing advanced AI capabilities with budget management. The upgraded console is available to all users immediately.

[00:11:49] And Google is introducing a new experimental feature called AI Mode in its search service, aiming to enhance the way users ask complex and multi-part questions. This mode, which is rolling out to Google One AI Premium subscribers, allows users to dig deeper into topics with advanced reasoning and multimodal capabilities.

[00:12:08] According to Robbie Stein, the vice president of product at Google Search, users are asking questions that are twice the length of traditional searches and following up with additional questions about a quarter of the time. The feature utilizes a technique called query fan out to conduct multiple related searches concurrently, providing users with comprehensive responses.

[00:12:29] Google emphasizes the importance of factuality, particularly in sensitive areas like health, and warns that while the system aims to be helpful, it may still make mistakes as it continues to develop. Why do we care? Let me give you some big picture reasons to care. The AI pricing wars are accelerating, making tools more affordable and widely used. Lower AI costs means greater accessibility, enabling more businesses to experiment with AI-powered data analysis and automation.

[00:12:59] Autonomous AI agents could revolutionize workflows, but governance and security risks remain high. Manus' AI error messages and inconsistencies suggest that fully autonomous AI is still very unreliable. If businesses over-rely on imperfect AI without oversight, it could create compliance risks, errors, and reputational damage. Microsoft's AI strategy is shifting, and businesses reliant on open AI should be prepared for future disruptions.

[00:13:27] Microsoft's exploration of Meta's and XAI's models suggests an AI model diversification and or commoditization trend, which could give businesses more AI options in the future. An AI democratization means more non-technical users will need support, creating opportunities for IT service firms. Businesses will need AI training, best practices, and security oversight, creating opportunities for providers to offer AI onboarding services.

[00:13:55] Are you ready to get your brand in front of the tech leaders shaping the future of managed services? Here at the Business of Tech, we offer flexible sponsorship opportunities to meet your needs, whether it's live show sponsorship, podcast advertising, event promotion, or custom webinars. From affordable exposure options to exclusive sponsorships, our offerings are designed to fit businesses and vendors of all sizes looking to make an impact.

[00:14:23] Prices start at just $500 per month, making our packages a fraction of typical event sponsorship costs. Be a part of the conversation that matters to IT service providers worldwide. Join us at MSP Radio and amplify your message where it counts. Visit MSP Radio dot com slash engage today to explore all the ways we can help you grow. Thanks for listening.

[00:14:52] Today is National Girl Scout Day, National Baked Scallops Day, and National Working Moms Day. Nerdy Ocon will be held in Palm Springs, California from April 7th through 9th. Visit NerdyOcon.com to learn all about it. The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech. If you've enjoyed the show, make sure you've subscribed or followed on your favorite platform.

[00:15:21] It's free and helps directly. Give us a review, too. If you want to support the show, visit patreon.com slash MSP Radio and you'll get access to content early. Or buy our Why Do We Care merch at businessof.tech. Have a question you want answered? We take listener questions, send them in, ideally as a voice memo or video to question at MSP Radio dot com. I answer listener questions live on our Wednesday live show on YouTube and LinkedIn.

[00:15:52] If you've got a comment or a thought on a story, put it in the comments if you're on YouTube or reach out on LinkedIn if you're listening to the podcast. And if you want to advertise on the show, visit MSP Radio dot com slash engage. Once again, thanks for listening and I will talk to you again on our next episode. Part of the MSP Radio Network.