loader from loading.io

Do AI Trust and Safety Measures Deserve to Fail?

The Cyberlaw Podcast

Release Date: 12/12/2023

World on the Brink with Dmitri Alperovitch show art World on the Brink with Dmitri Alperovitch

The Cyberlaw Podcast

Okay, yes, I promised to take a hiatus after episode 500. Yet here it is a week later, and I'm releasing episode 501. Here's my excuse. I read and liked Dmitri Alperovitch's book, "World on the Brink: How America Can Beat China in the Race for the 21st Century."  I told him I wanted to do an interview about it. Then the interview got pushed into late April because that's when the book is actually coming out. So sue me. I'm back on hiatus. The conversation  in the episode begins with Dmitri's background in cybersecurity and geopolitics, beginning with his emigration from the Soviet...

info_outline
Who’s the Bigger Cybersecurity Risk – Microsoft or Open Source? show art Who’s the Bigger Cybersecurity Risk – Microsoft or Open Source?

The Cyberlaw Podcast

There’s a whiff of Auld Lang Syne about episode 500 of the Cyberlaw Podcast, since after this it will be going on hiatus for some time and maybe forever. (Okay, there will be an interview with Dmitri Alperovich about his forthcoming book, but the news commentary is done for now.) Perhaps it’s appropriate, then, for our two lead stories to revive a theme from the 90s – who’s better, Microsoft or Linux? Sadly for both, the current debate is over who’s worse, at least for cybersecurity.   Microsoft’s sins against cybersecurity are laid bare in , Paul Rosenzweig reports. ...

info_outline
Taking AI Existential Risk Seriously show art Taking AI Existential Risk Seriously

The Cyberlaw Podcast

This episode is notable not just for cyberlaw commentary, but for its imminent disappearance from these pages and from podcast playlists everywhere.  Having promised to take stock of the podcast when it reached episode 500, I’ve decided that I, the podcast, and the listeners all deserve a break.  So I’ll be taking one after the next episode.  No final decisions have been made, so don’t delete your subscription, but don’t expect a new episode any time soon.  It’s been a great run, from the dawn of the podcast age, through the ad-fueled podcast boom, which I...

info_outline
The Fourth Antitrust Shoe Drops, on Apple This Time show art The Fourth Antitrust Shoe Drops, on Apple This Time

The Cyberlaw Podcast

The Biden administration has been aggressively pursuing antitrust cases against Silicon Valley giants like Amazon, Google, and Facebook. This week it was Apple’s turn. The Justice Department (joined by several state AGs)  filed a accusing Apple of improperly monopolizing the market for “performance smartphones.” The market definition will be a weakness for the government throughout the case, but the complaint does a good job of identifying ways in which Apple has built a moat around its business without an obvious benefit for its customers.  The complaint focuses on Apple’s...

info_outline
Social Speech and the Supreme Court show art Social Speech and the Supreme Court

The Cyberlaw Podcast

The Supreme Court is getting a heavy serving of first amendment social media cases. Gus Hurwitz covers two that made the news last week. In the , Justice Barrett spoke for a unanimous court in spelling out the very factbound rules that determine when a public official may use a platform’s tools to suppress critics posting on his or her social media page.  Gus and I agree that this might mean a lot of litigation, unless public officials wise up and simply follow the Court’s broad hint: If you don’t want your page to be treated as official, simply say up top that it isn’t official....

info_outline
Preventing Sales of Personal Data to Adversary Nations show art Preventing Sales of Personal Data to Adversary Nations

The Cyberlaw Podcast

This bonus episode of the Cyberlaw Podcast focuses on the national security implications of sensitive personal information. Sales of personal data have been largely unregulated as the growth of adtech has turned personal data into a widely traded commodity. This, in turn, has produced a variety of policy proposals – comprehensive privacy regulation, a weird proposal from Sen. Wyden (D-OR) to ensure that the US governments cannot buy such data while China and Russia can, and most recently an Executive Order to prohibit or restrict commercial transactions affording China, Russia, and...

info_outline
The National Cybersecurity Strategy – How Does it Look After a Year? show art The National Cybersecurity Strategy – How Does it Look After a Year?

The Cyberlaw Podcast

Kemba Walden and Stewart revisit the National Cybersecurity Strategy a year later. Sultan Meghji examines the ransomware attack on Change Healthcare and its consequences. Brandon Pugh reminds us that even large companies like Google are not immune to having their intellectual property stolen. The group conducts a thorough analysis of a "public option" model for AI development. Brandon discusses the latest developments in personal data and child online protection. Lastly, Stewart inquires about Kemba's new position at Paladin Global Institute, following her departure from the role of Acting...

info_outline
Regulating personal data for national security show art Regulating personal data for national security

The Cyberlaw Podcast

The United States is in the process of rolling out a for personal data transfers. But the rulemaking is getting limited attention because it targets transfers to our rivals in the new Cold War – China, Russia, and their allies. old office is drafting the rules, explains the history of the initiative, which stems from endless Committee on Foreign Investment in the United States efforts to impose such controls on a company-by-company basis. Now, with an as the foundation, the Department of Justice has published an that promises what could be years of slow-motion regulation. Faced with a...

info_outline
Are AI models learning to generalize? show art Are AI models learning to generalize?

The Cyberlaw Podcast

We begin this episode with describing major progress in conversions. Amazon flagged its new model as having “emergent” capabilities in handling what had been serious problems – things like speaking with emotion, or conveying foreign phrases. The key is the size of the training set, but Amazon was able to spot the point at which more data led to unexpected skills. This leads Paul and me to speculate that training AI models to perform certain tasks eventually leads the model to learn “generalization” of its skills. If so, the more we train AI on a variety of tasks – chat,...

info_outline
Death, Taxes, and Data Regulation show art Death, Taxes, and Data Regulation

The Cyberlaw Podcast

On the latest episode of The Cyberlaw Podcast, guest host Brian Fleming, along with panelists and discuss the latest U.S. government efforts to protect sensitive personal data, including the and the restricting certain bulk sensitive data flows to China and other countries of concern. Nate and Brian then discuss before the April expiration and debate what to make of a recent . Gus and Jane then talk about the , as well as , in an effort to understand some broader difficulties facing internet-based ad and subscription revenue models. Nate considers the implications of in its war against...

info_outline
 
More Episodes

It’s the last and probably longest Cyberlaw Podcast episode of 2023. To lead off, Megan Stifel takes us through a batch of stories about ways that AI, and especially AI trust and safety, manage to look remarkably fallible. Anthropic released a paper showing that race, gender, and age discrimination by AI models was real but could be dramatically reduced by instructing The Model to “really, really, really” avoid such discrimination. (Buried in the paper was the fact that the original, severe AI bias disfavored older white men, as did the residual bias that asking nicely didn’t eliminate.) Bottom line from Anthropic seems to be, “Our technology is a really cool toy, but don’t use if for anything that matters.”) In keeping with that theme, Google’s highly touted OpenAI competitor Gemini was release to mixed reviews when the model couldn’t correctly identify recent Oscar winners or a French word with six letters (it offered “amour”). The good news was for people who hate AI’s ham-handed political correctness; it turns out you can ask another AI model how to jailbreak your model, a request that can make the task go 25 times faster.

This could be the week that determines the fate of FISA section 702, David Kris reports. It looks as though two bills will go to the House floor, and only one will survive. Judiciary’s bill is a grudging renewal of 702 for a mere three years, full of procedures designed to cripple the program. The intelligence committee’s bill beats the FBI around the head and shoulders but preserves the core of 702. David and I explore the “queen of the hill” procedure that will allow members to vote for either bill, both, or none, and will send to the Senate the version that gets the most votes. 

Gus Hurwitz looks at the FTC’s last-ditch appeal to stop the Microsoft-Activision merger. The best case, he suspects, is that the appeal will be rejected without actually repudiating the pet theories of the FTC’s hipster antitrust lawyers.

Megan and I examine the latest HHS proposal to impose new cybersecurity requirements on hospitals. David, meanwhile, looks for possible motivations behind the FBI’s procedures for companies who want help in delaying SEC cyber incident disclosures. Then Megan and I consider the tough new UK rules for establishing the age of online porn consumers. I think they’ll hurt Pornhub’s litigation campaign against states trying to regulate children’s access to porn sites. 

The race to 5G is over, Gus notes, and it looks like even the winners lost. Faced with the threat of Chinese 5G domination and an industry sure that 5G was the key to the future, many companies and countries devoted massive investments to the technology, but it’s now widely deployed and no one sees much benefit. There is more than one lesson here for industrial policy and the unpredictable way technologies disseminate.

23andme gets some time in the barrel, with Megan and I both dissing its “lawyerly” response to a history of data breaches – namely changing its terms of service it harder for customers to sue for data breaches.

Gus reminds us that the Biden FCC only took office in that last month or two, and it is determined to catch up with the FTC in advancing foolish and doomed regulatory initiatives. This week’s example, remarkably, isn’t net neutrality. It’s worse. The Commission is building a sweeping regulatory structure on an obscure section of the 2021 infrastructure act that calls for the FCC to “facilitate equal access to broadband internet access service...”: Think we’re hyperventilating? Read Commissioner Brendan Carr’s eloquent takedown of the whole initiative. 

Senator Ron Wyden (D-OR) has a been in his bonnet over government access to smartphone notifications. Megan and I do our best to understand his concern and how seriously to take it. 

Wrapping up, Gus offers a quick take on Meta’s broadening attack on the constitutionality of the FTC’s current structure. David takes satisfaction from the Justice Department’s patient and successful pursuit of Russian Hacker Vladimir Dunaev for his role in creating TrickBot. Gus notes that South Korea’s law imposing internet costs on content providers is no match for the law of supply and demand.

Finally, in quick hits we cover: 

Download 485th Episode (mp3)

You can subscribe to The Cyberlaw Podcast using iTunes, Google Play, Spotify, Pocket Casts, or our RSS feed. As always, The Cyberlaw Podcast is open to feedback. Be sure to engage with @stewartbaker on Twitter. Send your questions, comments, and suggestions for topics or interviewees to [email protected]. Remember: If your suggested guest appears on the show, we will send you a highly coveted Cyberlaw Podcast mug! The views expressed in this podcast are those of the speakers and do not reflect the opinions of their institutions, clients, friends, families, or pets.