AI's Jagged Age: Memory Limits, Retrieval Bots, and Legal Battles Over Encryption and Privacy

[00:00:02] It's Friday, June 13th, 2025 and I'm Dave Sobel for Things to Know Today. AI models have a defined memory ceiling, reshaping the copyright debates. Sundar Pichai dubs this the Jagged Age of AI where brilliance meets breakdown. The rise of AI retrieval bots is rewriting the economics of online content. And MSP-centric security and service platforms are betting big on AI-enabled scale. Here's why these matter. This is the Business of Tech.

[00:00:33] Research from Meta, Google DeepMind, and Cornell and NVIDIA reveals that large language models have a fixed memorization capacity of approximately 8.6 bits per parameter. This finding is significant as it clarifies how much of the training data is memorized versus generalized, which is critical for understanding the operational mechanisms of these models and addressing copyright concerns.

[00:00:57] The study indicates that training on larger datasets does not increase the likelihood of memorizing specific data points. Instead, it distributes the memorization capacity across the dataset. For instance, a model trained on a dataset of 1.5 billion parameters can hold about 5.4 billion bits of information.

[00:01:18] The research employed a unique approach by training models on random bit strings to measure memorization without the influence of patterns found in natural language. The method provides a clearer understanding of how language models retain information, which is particularly relevant in ongoing discussions surrounding data privacy and copyright in AI.

[00:01:39] Google CEO Sundar Pichai has introduced a new term for the current phase of artificial intelligence, referring to it as artificial jagged intelligence. This concept highlights the nonlinear progress in AI development as researchers and developers face significant challenges despite achieving impressive milestones.

[00:01:59] Pichai explained that while advancements continue, many AI models still struggle with basic tasks, describing the current state of AI as marked by jagged edges, a mix of extraordinary capabilities and notable errors. This perspective echoes sentiments shared by Andres Karpathy, a prominent researcher in deep learning who emphasized the unpredictability of AI performance.

[00:02:23] A new breed of artificial intelligence bot is rapidly transforming how users access information online, with traffic from these retrieval bots increasing by 49% in the first quarter of this year compared to the previous quarter. Companies like OpenAI and Anthropic are deploying these bots to summarize content in real time, moving away from traditional search methods that provided links to multiple sources.

[00:02:46] According to data by Tolbit, a New York-based startup, the growth of retrieval bots is exponential, reflecting a rising demand for content, even as human traffic to news sites declines. Tolbit's CEO emphasized the need for publishers to adapt to this shift, stating that, quote, this is coming for everyone, end quote.

[00:03:07] The report notes that more than 26 million AI scrapes bypassed content blockers in March of 2025, highlighting the challenges publishers face in protecting their material. As landscape changes, producers of content may need to rethink how they engage with AI visitors to sustain their business models. Why do we care? Suna Pachai's framing of the current phase as jagged is more than a catchy term.

[00:03:33] It's an admission of AI's inconsistent maturity, astonishing performance in some areas, baffling failures in others. The rise of retrieval bots, from OpenAI and Anthropic especially, signals a massive redefinition of the web's economic structure. Users increasingly receive summarized outputs, not links. This has enormous implications for content-driven businesses. We're moving into a world where content is consumed by AI first, humans second.

[00:04:03] Businesses need to rethink attribution, tracking, and even monetization models accordingly. The AI landscape is coming less about building capability and more about governing, packaging, and protecting it. You're no longer just implementing tools, you're shaping AI outcomes, managing its risks, and helping clients find relevance in a transformed digital economy. Treat AI like electricity in a jagged power grid, valuable, unpredictable, and politically charged.

[00:04:31] How you wire it up for your clients, safely, sustainability, and strategically will determine your future value. 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.

[00:04:58] From affordable exposure options to exclusive sponsorships, our offerings are designed to fit businesses and vendors of all sizes looking to make an impact. 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.

[00:05:27] Visit MSP Radio dot com slash engage today to explore all the ways we can help you grow. Two stories I've reported on before that there are updates to. WhatsApp has announced its intention to intervene in a legal case between Apple and the British government over access to iCloud accounts, which raises significant privacy concerns.

[00:05:50] Will Carthcott, the chief of WhatsApp, emphasized that the meta-owned messaging app aims to protect global privacy rights, stating that the UK's secret order could set a dangerous precedent for weakening encryption. The British government had issued a technical capability notice to Apple, demanding the halt of an end-to-end encryption feature that would prevent authorities from accessing iCloud data, even with a warrant.

[00:06:13] The Investigatory Powers Tribunal recently confirmed that Apple is suing the government regarding this legal order, which has sparked debate about transparency and privacy implications in the access to encrypted messaging. The U.S. Cybersecurity and Infrastructure Security Agency has lost another senior leader, with Executive Director Bridget Bean announcing her retirement after three and a half years with the agency.

[00:06:37] Bean served as the acting head of the agency for five months following the departure of former director Jen Easterly and has been recognized for her significant contributions to the agency's operational effectiveness. Bean's exit comes amid a broader talent exodus from the agency, where nearly one-third of the staff have reportedly left since the Trump administration took office. As of May 30th, five of the six operational division heads and six of the ten regional office leaders had departed.

[00:07:06] The agency is currently without a confirmed director, as the nomination of Sean Plankey is pending a full Senate vote with opposition from Senator Ron Wyden unless the agency releases a report on telecommunications security. The agency is also facing potential budget cuts, with proposals ranging from $135 million to $495 million, further complicating its leadership and operational challenges. Why do we care?

[00:07:33] WhatsApp decision to intervene in Apple's legal battle with the government highlights a critical and growing pattern. Platforms are no longer passive actors in surveillance debates. With the UK issuing secret orders that aim to halt end-to-end encryption, we're seeing a direct clash between national security priorities and global privacy standards. The stakes here are high. If governments succeed in forcing platforms to weaken encryption, even behind closed doors, it sets a precedent that could cascade across jurisdictions.

[00:08:02] Now, some may argue the sky isn't falling. Encryption policy debates are cyclical, and CISA, like many federal agencies, has weathered leadership shifts before. Apple, Meta, and others are fighting back in court, a sign of a functioning legal check. And while CISA is short on confirmed leaders, much of its operational staff remains intact. That said, uncertainty is the operative word, and in cybersecurity, that's not where you want to be.

[00:08:31] Arctic Wolf has enhanced its managed service provider partner program and launched Aurora Endpoint Security for MSPs. The redesigned program offers partners scalable pricing and simplified deal structures, which are critical as the MSP market expands. The Aurora Endpoint Security leverages insights from over 10,000 customers and processes more than 8 trillion security observations weekly, allowing for improved detection and response while significantly reducing alert noise.

[00:08:58] Shield Technology Partners has launched an AI-enabled managed IT service platform, marking a significant development in the technology services sector. With initial funding exceeding $100 million, the platform aims to provide local businesses with access to national-scale resources and expertise. The market for IT services is currently valued at over $700 billion and is evolving rapidly.

[00:09:22] The platform will collaborate with foundational partners such as ClearFuse Networks and IronOrbit, allowing them to enhance their operations while maintaining their unique offerings. Microsoft Edge has introduced a new feature called Secure Password Deployment, aimed at enhancing password management for businesses.

[00:09:39] This feature allows administrators to deploy encrypted shared passwords to users within their organization, minimizing the risk of unauthorized access and reducing the likelihood of employees sharing passwords through insecure methods like sticky notes or emails.

[00:09:54] Available for Microsoft 365 Business Premium, E3 and E5 subscriptions, the Secure Password Deployment feature integrates with the Microsoft Edge Management Service, enabling administrators to configure browser settings and deploy passwords to specific user groups seamlessly. According to Microsoft, this integration not only aligns with zero-trust principles, but also ensures compliance with organizational policies without requiring manual key management.

[00:10:22] Passwords are encrypted using Microsoft's information protection SDK and are automatically available in users' Edge work profiles on managed Windows devices ready for autofill on designated websites. WiseHire has launched an AI-powered recruiting assistant called WiseHire Scout, designed to streamline the hiring processes for small businesses. The tool automates various recruiting tasks, allowing employers to identify and connect with top candidates more efficiently.

[00:10:51] The Houston-based hiring platform announced that WiseHire Scout is available to all current and future customers at no additional cost. Early users have reported that the assistant has helped reduce time to hire by up to 90%, while also rediscovering qualified candidates from previous roles. Why do we care? The IT services industry is moving into a phase where platform gravity, operational design, and trust in automation determine competitive edge.

[00:11:19] Across these announcements, AI is used as a given, not a selling point. That is telling. Whether it's hiring, cybersecurity, IT support, or password management, AI is the expected layer. The differentiator now is execution, interoperability, and trust. Vendors and IT providers alike should stop leading with, we use AI, and start focusing on how it reduces noise, enables scale, or drives better outcomes. These stories aren't about new tech, they're about who helps the customer make it work.

[00:11:50] That's your opening. Ending the week with some big ideas. In Business Insider, a consultant who assists law firms in selecting software warns that the legal technology sector may be facing an impending bubble. Zach Abramowitz, founder of Killer Whale Strategies, noted that while many law firms are currently testing various legal tech products, this surge in revenue could be misleading, as most of these trials are pilot programs rather than long-term commitments.

[00:12:18] With a significant increase in funding and the emergence of successful startups, the legal tech market is attracting considerable attention. However, Abramowitz cautions that in the next 12 to 18 months, firms will consolidate their choices around a few top-performing technologies, leaving less successful vendors at risk of declining revenues.

[00:12:38] Additionally, Rick Zullo, a software investor, highlighted concerns that if all companies in the sector are performing well, it could indicate a lack of competitive differentiation, potentially signaling trouble ahead for the industry. In Information Week, Matt Aslett, director of research at the technology advisory firm Information Services Group, warns that users should be vigilant, as AI models are designed to replicate content rather than ensure factual accuracy.

[00:13:08] Aslett emphasizes the importance of verifying the output generated by generative AI, which lacks semantic understanding and can produce misleading information. Documented instances of generative AI errors include chatbots providing incorrect customer information, and lawyers facing penalties for submitting filing with fabricated legal cases.

[00:13:27] To mitigate these risks, enterprises are encouraged to adopt comprehensive validation processes and governance frameworks to help identify and correct inaccuracies in generative AI outputs. In light of the challenges, experts suggest improving generative AI accuracy through methods such as specialized training and prompt engineering, while also advocating for automated reasoning to enhance correctness in generated content.

[00:13:54] And it's fun to revisit the idea of AI in the drive-thru. Fast food chains are rapidly modernizing their drive-thru services with artificial intelligence, yet a recent survey reveals that most Americans still prefer ordering from a human. According to a YouGov survey conducted in January of 2025, 55% of respondents expressed a preference for speaking to a person at the drive-thru, while only 4% indicated they would rather use an automated system.

[00:14:21] While companies like Wendy's and Taco Bell are advancing their AI initiatives, customer trust in these technologies remains low. For instance, McDonald's recently ended its AI drive-thru trial after receiving negative feedback regarding order accuracy and customer experience. Despite the push-thru trial after using the data, many customers still value the personal interaction that comes with ordering from a human being. Why do we care?

[00:14:47] Well, the legal sector's experimentation with tech mirrors what we've seen across industries, a rush to try tools without long-term commitment. The surge in vendor activity, investor enthusiasm, and short-term revenue might feel like traction, but Abramowitz's warning points to a coming correction. The consolidation phase he predicts is where many vendors hit a wall, and so will the MSPs and IT consultants who backed the wrong horse.

[00:15:12] Aslitt's warning about generative AI hallucinations isn't new, but his emphasis on systematic governance deserves amplification. If your clients are experimenting with ChatGPT, co-pilot, or legal tech tools powered by generative AI, you can no longer ignore validation, risk management, and prompt engineering as core service competencies. Despite the push toward automation in fast food, consumer behavior isn't keeping pace. The McDonald's AI trial failure and survey data point to a deeper truth.

[00:15:41] Just because a technology is available doesn't mean it's welcome or appropriate. It's a crucial reminder for IT providers advising clients on automation. Customer experience isn't just about efficiency. If AI reduces perceived quality, trust, or comfort, especially in service-centric businesses, expect backlash. For IT service providers, these stories are a clear directive. Don't chase the hype. Guide the adoption.

[00:16:08] Legal tech, generative AI, and AI-enabled CX all have promise, but only if deployed with strategic clarity, operational validation, and an understanding of customer psychology. The winners won't be those who sell the most tools. They'll be the ones who help clients make the right bets and back them with resilient, human-aware service models. Thanks for listening.

[00:16:32] Today is National Sewing Machine Day, International Axe Throwing Day, and National Blame Someone Else Day. The last one just seems wrong. Join me for a webinar sponsored by Nerdio, Modern Endpoint Management with Intune, What Works and What Doesn't. Bit.ly slash Nerdio webinar to sign up or link in the show notes. The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech.

[00:17:02] If you've enjoyed the show, make sure you've subscribed or followed on your favorite platform. 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?

[00:17:25] We take listener questions, send them in, ideally as a voice memo or video to question at MSP radio.com. I answer listener questions live on our Wednesday live show on YouTube and LinkedIn. 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. I will talk to you again on our next episode.

[00:17:57] Part of the MSP radio network. Do you believe that? Have a great day. But we won't have a chance to see what does not happen.