Skating on Stilts

479. The new AI executive order, plus Scott Shapiro on "Fancy Bear Goes Phishing"

In this episode of the Cyberlaw Podcast, I take advantage of Scott Shapiro’s participation to interview him about his book, Fancy Bear Goes Phishing – The Dark History of the Information Age, in Five Extraordinary Hacks. It’s a remarkable tutorial on cybersecurity, told through stories that you may think you've already heard until you see what Scott has turned up by digging into historical and legal records. We cover the Morris worm, the Paris Hilton hack, and the earliest Bulgarian virus writer’s nemesis. Along the way, we share views about the refreshing emergence of a well-paid profession largely free of the credentialism that infects so much of the American economy. In keeping with the rest of the episode, I ask Bing Image Creator to generate alternative artwork for the book.

In the news roundup, Michael Ellis walks us through the “sweeping” ™ White House executive order on artificial intelligence. The tl;dr: the order may or may not actually have real impact on the field. The same can probably be said of the advice now being dispensed by AI’s “godfathers.”™ -- the keepers of the flame for AI existential risk who have urged that AI companies devote a third of their R&D budgets to AI safety and security and accept liability for serious harm. Scott and I puzzle over how dangerous AI can be when even the most advanced engines can only do multiplication successfully 85% of the time. Along the way, we evaluate methods for poisoning training data and their utility for helping starving artists get paid when their work is repurposed by AI.

Speaking of AI regulation, Nick Weaver offers a real-life example: the California DMV’s immediate suspension of Cruise’s robotaxi permit after a serious accident that the company handled poorly.

Michael talks about what’s been happening in the Google antitrust trial, to the extent that anyone can tell, thanks to the heavy confidentiality restrictions imposed by Judge Mehta. One number that escaped -- $26 billion in payments to maintain Google as everyone’s default search engine – draws plenty of commentary.

Scott and I try to make sense of CISA’s claim that its vulnerability list has produced cybersecurity dividends. We are inclined to agree that there’s a pony in there somewhere.

Nick explains why it’s dangerous to try to spy on Kaspersky. The rewards may be big, but so is the risk that your intelligence service will be pantsed. Nick notes that using Let’s Encrypt as part of your man in the middle attack has risks as well – advice he probably should deliver auf Deutsch.

Scott and I cover a great Andy Greenberg story about a team of hackers who discovered how to unlock a vast store of bitcoin  but may not see a payoff soon. I reveal my connection to the story.

Michael and I share thoughts about the effort to renew section 702 of FISA, which lost momentum during the long battle over choosing a Speaker of the House. I note that USTR has surrendered to reality in global digital trade and point out that last week’s story about judicial interest in tort cases against social media turned out to be the first robin in what now looks like a remake of The Birds.

Download 479th 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

478. Administration fails forward on China chip exports

This episode of the Cyberlaw Podcast begins with the administration's aggressive new rules on chip exports to China. Practically every aspect of the rules it announced just eight months ago was sharply tightened, Nate Jones reports. The changes are so severe, I suggest, that they make the original rules look like a failure that had to be overhauled to work.

Much the same could be said about the Biden administration's plan for an executive order on AI regulation that Chessie Lockhart thinks will  focus on government purchases. As a symbolic expression of best AI practice, procurement-focused rules make symbolic sense. But given the current government market for AI, it's hard to see them having much bite. So look for more in the way of teeth down the road as the regulatory process again fails forward.

If it's regulatory bite you want, Nate says, the EU has now sketched out what appears to be version 3.0 of its AI Act. It doesn't look all that much like Versions 1.0 or 2.0, but it's sure to take the world by storm, fans of the Brussels Effect tell us. I note that the new version includes plans for fee-driven enforcement and suggest that the scope of the rules is already being tailored to ensure fee revenue from popular but not especially risky AI models.

Jane Bambauer offers a kind review of  Marc Andreessen's "'Techno-Optimist Manifesto".  We both end up agreeing more than we disagree with Marc's arguments, if not his bombast -- a style that I suspect owes much to extreme mountaineering.

Chessie reveals the Achilles heel of a growing state movement to require that registered data brokers delete personal data on request. It turns out that a lot of the data brokers, just aren't registering.

The Supreme Court, moving with surprising speed at the Solicitor General's behest, has granted cert and a stay  in the social media jawboning case, which was brought by Missouri among other states to stop federal agencies from demanding that social media suppress speech the federal government disagrees with. I note that the SG's desperation to win this case has led it to make surprisingly creative arguments, as illustrated in yet another Cybertoonz explainer.

Social media's loss of public esteem may be showing up in judicial decisions. Jane reports on a California decision allowing a negligence lawsuit to go forward against kids' social media for marketing an addictive product. I'm happier than Jane to see that the bloom is off the section 230 rose, but we agree that suing companies for making their product's too attractive may run into a few pitfalls on the way to judgment. Listeners who don't remember the Reagan administration may benefit from my short history of the California judge who wrote the opinion.

And speaking of tort liability for tech products, Chessie tells us that Chinny Sharma, another Cyberlaw podcast stalwart, has an article in Lawfare confessing some fondness for products liability (as opposed to negligence) lawsuits for cybersecurity failures.

Chessie also breaks down a Colorado Supreme Court decision approving a keyword search for an arson-murder suspect. Although played as a win for keyword searches in the press, it's actually a loss. The search results were deemed admissible only because the government's good faith excused what the court considered its lack of probable cause. I award EFF the "sore winner" award for its whiny screed complaining that, while the court handed EFF a victory on the impropriety of the search, the court didn't also give a get-out-of-jail-free card to the scumbags accused of burning five people to death.

Finally,  Nate and I explain why the Cybersecurity and Infrastructure Security Agency shouldn't expect Congress to pass what used to be a yearly batch of routine small-ball cyber bills. CISA overplayed its hand in the misinformation wars over the  2020 election, going so far as to consider curbs on "malinformation" – information that is true but inconvenient for the government. This has led a lot of conservatives to look for reasons to cut CISA's budget. Sen. Rand Paul (R-KY) gets special billing.

Download 478th 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.

Cybertoonz visits Murthy v. Missouri

The effort by Missouri to enjoin the Biden administration's "jawboning" of social media companies has reached the Supreme Court. The Solicitor General filed a long brief seeking a stay of the Fifth Circuit's injunction and explaining why cert should be granted. The Court granted cert and the stay (the latter over three dissents).

The SG's brief was remarkable for its innovative approach to the legal issues in the case, first proclaiming that states like Missouri have no first amendment right to hear from their citizens and then having to resurrect something very like a first amendment right for the federal government to speak to its citizens. Confused?  Never fear. Where technology leads lawyers and policymakers to new heights of absurdity, Cybertoonz will be there to explain it all.

 

477. Will CISOs have to choose between getting rich or going to jail?

This episode of the Cyberlaw Podcast delves into a False Claims Act lawsuit against Penn State University by a former CIO to one of its research units. The lawsuit alleges that Penn State faked security documents in filings with the Defense Department. Because it’s a so-called qui tam case, Tyler Evans explains, the plaintiff could recover a portion of any funds repaid by Penn State. Which is preferable to the alternative: If the employee was complicit in a scheme to mislead DoD, the False Claims Act isn’t limited to civil cases like this one; the Justice Department can pursue criminal sanctions too – although Tyler notes that, so far, Justice has been slow to take that step.

In other news, Jeffery Atik and I try to make sense of a New York Times story about Chinese bitcoin miners setting up shop near a Microsoft data center and a DoD base. The reporter seems sure that the Chinese miners are doing something suspicious, but it’s never clear exactly what the problem is.

California Governor Gavin Newsom (D) is widely believed to be positioning himself for a Presidential run, maybe as early as next year. In that effort, he’s been able to milk the Sacramento Effect, in which California adopts legislation that more or less requires the country to follow its lead. One such law is the DELETE (Data Elimination and Limiting Extensive Tracking and Exchange) Act, which, Jim Dempsey reports, would require all data brokers to delete the personal data of anyone who makes a request to a centralized California agency. This will be bad news for most data brokers, and good news for the biggest digital ad companies like Google and Amazon, since those companies acquire data directly from their customers and not through purchase.

Another California law that could have similar national impact bans social media from “aiding or abetting” child abuse. This framing is borrowed from FOSTA (Allow States and Victims to Fight Online Sex Trafficking Act)/SESTA (Stop Enabling Sex Traffickers Act), a federal law that prohibited aiding and abetting sex trafficking and led to the demise of sex classified ads and the publications they supported around the country.

I cover the overdetermined collapse of EPA’s effort to impose cybersecurity regulation on the nation’s water systems. I predict we won’t see an improvement in water system cybersecurity without new legislation.

Justin lays out how badly the Senate is fracturing over regulation of AI. Jeffery and I puzzle over the Commerce Department’s decision to allow South Korean DRAM makers to keep using U.S. technology in their Chinese foundries.

Jim lays out the unedifying history of Congressional and administration efforts to bring a hammer down on TikTok while Jeffery evaluates the prospects for Utah’s lawsuit against TikTok based on a claim that the  app has a harmful impact on children.

Finally, in what looks like good news about AI transparency, Jeffery covers Anthropic’s research showing that – sometimes – it’s possible to identify the features that an AI model is relying upon, showing how the model weights features like legal talk or reliance on spreadsheet data. It’s a long way from there to full explanations of how the model makes its decisions, but Anthropic thinks we’ve moved from needing more science to needing more engineering. (Credit as always to Bing Image Creator for the graphics.)

Download 477th 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.

476. The PCLOB report on 702 -- we interview both sides

The debate over section 702 of FISA is heating up as the end-of-year deadline for reauthorization draws near. The debate is reflected in a report from the Privacy and Civil Liberties Oversight Board. That report was not unanimous.

In the interest of helping listeners understand the report and its recommendations, the Cyberlaw Podcast has produced a bonus episode 476, featuring two board members who represent the board's divergent views --  Beth Williams, a Republican-appointed member, and Travis LeBlanc, a Democrat-appointed member.

It's a great introduction to the 702 program, touching first on the very substantial points of agreement about it and then on the concerns and recommendations for addressing those concerns.

Best of all, the conversation ends with a surprise consensus on the importance of using the program to vet travelers to the United States and holders of security clearances.

Download 476th Episode (mp3)

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.

 

 

475. Technology and Terror

Today's episode of the Cyberlaw Podcast begins, as it must, with Saturday's appalling Hamas attack on Israeli civilians. I ask Adam Hickey and Paul Rosenzweig, both with long histories in counterterrorism, to comment on the attack and what lessons the U.S. should draw from it, whether in terms of revitalized intelligence programs or the need for workable defenses against drone attacks.

In other news, Adam covers the disturbing prediction that the U.S. and China have a fifty percent chance of armed conflict in the next five years – and the supply chain consequences of increasing conflict. Meanwhile, Western companies who were hoping to sit the conflict out may not be given the chance. Adam also covers the related EU effort to assess risks posed by four key technologies.

Paul and I share our doubts about the Red Cross's effort to impose ethical guidelines on hacktivists in war. Not that we needed to; the hacktivists seem perfectly capable of expressing their doubts on their own.

The Fifth Circuit has expanded its injunction against the U.S. government, prohibiting the White House and several agencies from encouraging or coercing social media to suppress "disinformation." Adam, who oversaw FBI efforts to counter foreign disinformation, takes a different view of the facts than the Fifth Circuit. In the same vein, we note a recent paper from two former Facebook content moderators who say that government jawboning of social media really does work (as if you had any doubts).

Paul comments on the EU vulnerability disclosure proposal and the hostile reaction it has attracted from some sensible people.

Adam and I find value in an op-ed that explains the camps locked in a weird war, not over whether to regulate AI but over how and why.

And, finally, Paul mourns yet another step in Apple's step-by-step surrender to Chinese censorship and social control.

Download 475th 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.

 

Like all the big AI companies, Bing's Image Creator software has a content policy that prohibits creation of images that encourage sexual abuse, suicide, graphic violence, hate speech, bullying, deception, and disinformation. Some of the rules are heavy-handed even by the usual "trust and safety" standards (hate speech is defined as speech that "excludes" individuals on the basis of any actual or perceived "characteristic that is consistently associated with systemic prejudice or marginalization"). Predictably, this will exclude a lot of perfectly anodyne images. But the rules are the least of it. The more impactful, and interesting, question is how those rules are actually applied.

I now have a pinhole view of AI safety rules in action, and it sure looks as though Bing is taking very broad rules and training their engine to apply them even more broadly than anyone would expect.

Here's my experience. I have been using Bing Image Creator lately to create Cybertoonz (examples here, here, and here), despite my profound lack of artistic talent. It had the usual technical problems -- too many fingers, weird faces -- and some problems I suspected were designed to avoid "gotcha" claims of bias. For example, if I asked for a picture of members of the European Court of Justice, the engine almost always created images of more women and identifiable minorities than the CJEU is likely to have in the next fifty years. But if the AI engine's political correctness detracted from the message of the cartoon, it was easy enough to prompt for male judges, and Bing didn't treat this as "excluding" images by gender, as one might have feared.

My more recent experience is a little more disturbing. I created this Cybertoonz cartoon to illustrate Silicon Valley's counterintuitive claim  that social media is engaged in protected speech when it suppresses the speech of many of its users. My image prompt was some variant of "Low angle shot of a male authority figure in a black t-shirt who stands and speaks into a loudspeaker in a large group of seated people wearing gags or tape over their mouths.  Digital art lo-fi".

As always, Bing's first attempt was surprisingly good, but flawed, and getting a useable version required dozens of edits of the prompt. None of the images were quite right.  I finally settled for the one that worked best, turned it into a Cybertoonz cartoon, and published it.  But I hadn't given up on finding something better, so I went back the next day and ran the prompt again.

This time, Bing balked. It told me my prompt violated Bing's safety standards:

After some experimenting, it became clear that what Bing objected to was depicting an audience "wearing gags or tape over their mouths."

How does this violate Bing's safety rules? Are gags an incitement to violence? A marker for "[n]on-consensual intimate activity"? In context, those interpretations of the rules are ridiculous. But Bing isn't interpreting the rules in context. It's trying to write additional code to make sure there are no violations of the rules, come hell or high water. So if there's a chance that the image it produces might show non-consensual sex or violence, the trust and safety code is going to reject it.

This is almost certainly the future of AI trust and safety limits. It will start with overbroad rules written to satisfy left-leaning critics of Silicon Valley. Then those overbroad rules will be further broadened by hidden code written to block many perfectly compliant prompts just to ensure that it blocks a handful of noncompliant prompts.

In the Cybertoonz context, such limits on AI output are simply an annoyance. But AI isn't always going to be a toy.  It's going to be used in medicine, hiring, and other critical contexts, and the same dynamic will be at work there. AI companies will be pressured to adopt trust and safety standards and implementing code that aggressively bar outcomes that might offend the left half of American political discourse. In applications that affect people's lives, however, the code that ensures those results will have a host of unanticipated consequences, many of which no one can defend.

Given the stakes, my question is simple. How do we avoid those consequences, and who is working to prevent them?

Cybertoonz explains Netchoice v. Paxton

One of the biggest Supreme Court cases this year will be a Big Tech challenge to Texas and Florida laws that seek to impose limits and transparency on social media content regulation.  Silicon Valley argues that these laws interfere with Big Tech's first amendment right to "cull and curate" what other people say on their platforms. The Biden administration agrees, arguing in its brief that deciding what users can and cannot say is not censorship but the constitutionally protected exercise of editorial judgment: “The act of culling and curating the content that users see is inherently expressive, even if the speech that is collected is almost wholly provided by users.”

As a public service, Cybertoonz has boiled the argument down even further:

 

474. Is silencing a few million Americans a form of protected speech?

The Supreme Court has granted certiorari to review two big state laws trying to impose limits on social media censorship (or "curation," if you prefer) of platform content. Paul Stephan and I spar over the right outcome, and the likely vote count, in the two cases. One surprise: we both think that the platforms' claim of a first amendment "right to curate" is in tension with their claim that they, uniquely among speakers, should have an immunity for that form of speech.

Maury weighs in to note that the EU is now gearing up to bring social media to heel on the "disinformation" front. That fight will be ugly for Big Tech, he points out, because Europe doesn't care if it puts social media out of business, since it's an American industry. I point out that elites all across the globe have rallied to meet and defeat social media's challenge to their agenda-setting and reality-defining authority. India is aggressively doing the same.

Paul covers another big story in law and technology: The FTC has sued Amazon for antitrust violations – essentially price gouging and tying. Whether the conduct alleged in the complaint is even a bad thing will depend on the facts found by the court, so the case will be hard fought. And, given the FTC's track record, no one should be betting against Amazon.

Nick Weaver explains the dynamic behind the massive MGM and Caesars hacks. As with so many globalized industries, the ransomware supply chain now has Americans in marketing (or social engineering, if you prefer) and foreign technology suppliers. Nick thinks it's time to OFAC 'em all.

Maury explains the latest bulk intercept decision from the European Court of Human Rights. The UK has lost again, but it's not clear how much difference that will make. The ruling says that non-Brits can sue the UK over bulk interception, but the court has already made clear that, with a few legislative tweaks, bulk interception is legal under the European human rights convention.

More bad news for 230 maximalists: it turns out that Facebook can be sued for allowing advertisers to target ads based on age and gender. The platform lost its immunity because it facilitated advertiser's allegedly discriminatory targeting.

The UK competition authorities are seeking greater access to AI's inner workings to assess risks, but Maury Shenk is sure this is part of a light touch on AI regulation that is meant to make the UK a safe European harbor for AI companies.

In a few quick hits and updates:

• I explain the splintered PCLOB report, which endorses 702 renewal, with widely diverging proposals for reform.
• Paul tells us that the Biden Administration plans to bring back "net neutrality" rules. Hey, if we're going to start reviving hits from 2010, can we bring back Ke$ha's TikTok instead?
• I flag an issue likely to spark a surprisingly bitter clash between the administration and cloud providers – Know Your Customer rules. The government thinks it's irresponsible from a cybersecurity point of view to let randos spin up virtual machines. The industry doesn't think the market will tolerate any other way of doing business.
• Speaking of government-industry clashes, it looks like Apple is caught between (a) Chinese demands that it impose tough new controls on the apps it makes available in its app store and (b) human decency. Maury has the story. And I've got a solution. Apple should just rebrand the totalitarian new controls it must adopt as "app curation." Seems to be working for everyone else.

Download 474th 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.

 

 

The President's Civil Liberties Oversight Board's report on section 702 of FISA

I summarize the President's Civil Liberties Oversight Board (PCLOB) report on section 702 of FISA in this Lawfare article.  Quick summary:

"The PCLOB report is a gold mine of authoritative information about Section 702, and evaluating the recommendations is a good way to refine one’s view of what reforms are needed. Whether the report will have much impact on the debate over renewal, however, is less clear. The unanimous support for renewal may be influential in the sense that it confirms a sentiment that already seems widespread in Congress, despite the FBI’s travails. The report’s inability to agree on more than that will dissipate its influence, particularly because understanding the dueling proposals for 702 reform requires working through hundreds of dense pages."