Page 4

Upcoming Speaking Engagements

This is a current list of where and when I am scheduled to speak:

  • My coauthor Nathan E. Sanders and I are speaking at the Rayburn House Office Building in Washington, DC at noon ET on November 17, 2025. The event is hosted by the POPVOX Foundation and the topic is “AI and Congress: Practical Steps to Govern and Prepare.”
  • I’m speaking on “Integrity and Trustworthy AI” at North Hennepin Community College in Brooklyn Park, Minnesota, USA, on Friday, November 21, 2025, at 2:00 PM CT. The event is cohosted by the college and The Twin Cities IEEE Computer Society.
  • Nathan E. Sanders and I will be speaking at the MIT Museum in Cambridge, Massachusetts, USA, on December 1, 2025, at 6:00 pm ET.
  • Nathan E. Sanders and I will be speaking at a virtual event hosted by City Lights on the Zoom platform, on December 3, 2025, at 6:00 PM PT.
  • I’m speaking and signing books at the Chicago Public Library in Chicago, Illinois, USA, on February 5, 2026. Details to come.

The list is maintained on this page.

Tags:

Posted on November 14, 2025 at 12:08 PM • View Comments

The Role of Humans in an AI-Powered World

As AI capabilities grow, we must delineate the roles that should remain exclusively human. The line seems to be between fact-based decisions and judgment-based decisions.

For example, in a medical context, if an AI was demonstrably better at reading a test result and diagnosing cancer than a human, you would take the AI in a second. You want the more accurate tool. But justice is harder because justice is inherently a human quality in a way that “Is this tumor cancerous?” is not. That’s a fact-based question. “What’s the right thing to do here?” is a human-based question.

Chess provides a useful analogy for this evolution. For most of history, humans were best. Then, in the 1990s, Deep Blue beat the best human. For a while after that, a good human paired with a good computer could beat either one alone. But a few years ago, that changed again, and now the best computer simply wins. There will be an intermediate period for many applications where the human-AI combination is optimal, but eventually, for fact-based tasks, the best AI will likely surpass both.

The enduring role for humans lies in making judgments, especially when values come into conflict. What is the proper immigration policy? There is no single “right” answer; it’s a matter of feelings, values, and what we as a society hold dear. A lot of societal governance is about resolving conflicts between people’s rights—my right to play my music versus your right to have quiet. There’s no factual answer there. We can imagine machines will help; perhaps once we humans figure out the rules, the machines can do the implementing and kick the hard cases back to us. But the fundamental value judgments will likely remain our domain.

This essay originally appeared in IVY.

Tags: ,

Posted on November 14, 2025 at 7:00 AM • View Comments

Book Review: The Business of Secrets

The Business of Secrets: Adventures in Selling Encryption Around the World by Fred Kinch (May 24, 2024)

From the vantage point of today, it’s surreal reading about the commercial cryptography business in the 1970s. Nobody knew anything. The manufacturers didn’t know whether the cryptography they sold was any good. The customers didn’t know whether the crypto they bought was any good. Everyone pretended to know, thought they knew, or knew better than to even try to know.

The Business of Secrets is the self-published memoirs of Fred Kinch. He was founder and vice president of—mostly sales—at a US cryptographic hardware company called Datotek, from company’s founding in 1969 until 1982. It’s mostly a disjointed collection of stories about the difficulties of selling to governments worldwide, along with descriptions of the highs and (mostly) lows of foreign airlines, foreign hotels, and foreign travel in general. But it’s also about encryption.

Datotek sold cryptographic equipment in the era after rotor machines and before modern academic cryptography. The company initially marketed computer-file encryption, but pivoted to link encryption—low-speed data, voice, fax—because that’s what the market wanted.

These were the years where the NSA hired anyone promising in the field, and routinely classified—and thereby blocked—publication of academic mathematics papers of those they didn’t hire. They controlled the fielding of strong cryptography by aggressively using the International Traffic in Arms regulation. Kinch talks about the difficulties in getting an expert license for Datotek’s products; he didn’t know that the only reason he ever got that license was because the NSA was able to break his company’s stuff. He had no idea that his largest competitor, the Swiss company Crypto AG, was owned and controlled by the CIA and its West German equivalent. “Wouldn’t that have made our life easier if we had known that back in the 1970s?” Yes, it would. But no one knew.

Glimmers of the clandestine world peek out of the book. Countries like France ask detailed tech questions, borrow or buy a couple of units for “evaluation,” and then disappear again. Did they break the encryption? Did they just want to see what their adversaries were using? No one at Datotek knew.

Kinch “carried the key generator logic diagrams and schematics” with him—even today, it’s good practice not to rely on their secrecy for security—but the details seem laughably insecure: four linear shift registers of 29, 23, 13, and 7 bits, variable stepping, and a small nonlinear final transformation. The NSA probably used this as a challenge to its new hires. But Datotek didn’t know that, at the time.

Kinch writes: “The strength of the cryptography had to be accepted on trust and only on trust.” Yes, but it’s so, so weird to read about it in practice. Kinch demonstrated the security of his telephone encryptors by hooking a pair of them up and having people listen to the encrypted voice. It’s rather like demonstrating the safety of a food additive by showing that someone doesn’t immediately fall over dead after eating it. (In one absolutely bizarre anecdote, an Argentine sergeant with a “hearing defect” could understand the scrambled analog voice. Datotek fixed its security, but only offered the upgrade to the Argentines, because no one else complained. As I said, no one knew anything.)

In his postscript, he writes that even if the NSA could break Datotek’s products, they were “vastly superior to what [his customers] had used previously.” Given that the previous devices were electromechanical rotor machines, and that his primary competition was a CIA-run operation, he’s probably right. But even today, we know nothing about any other country’s cryptanalytic capabilities during those decades.

A lot of this book has a “you had to be there” vibe. And it’s mostly tone-deaf. There is no real acknowledgment of the human-rights-abusing countries on Datotek’s customer list, and how their products might have assisted those governments. But it’s a fascinating artifact of an era before commercial cryptography went mainstream, before academic cryptography became approved for US classified data, before those of us outside the triple fences of the NSA understood the mathematics of cryptography.

This book review originally appeared in AFIO.

Tags: , , ,

Posted on November 13, 2025 at 7:09 AM • View Comments

On Hacking Back

Former DoJ attorney John Carlin writes about hackback, which he defines thus: “A hack back is a type of cyber response that incorporates a counterattack designed to proactively engage with, disable, or collect evidence about an attacker. Although hack backs can take on various forms, they are—­by definition­—not passive defensive measures.”

His conclusion:

As the law currently stands, specific forms of purely defense measures are authorized so long as they affect only the victim’s system or data.

At the other end of the spectrum, offensive measures that involve accessing or otherwise causing damage or loss to the hacker’s systems are likely prohibited, absent government oversight or authorization. And even then parties should proceed with caution in light of the heightened risks of misattribution, collateral damage, and retaliation.

As for the broad range of other hack back tactics that fall in the middle of active defense and offensive measures, private parties should continue to engage in these tactics only with government oversight or authorization. These measures exist within a legal gray area and would likely benefit from amendments to the CFAA and CISA that clarify and carve out the parameters of authorization for specific self-defense measures. But in the absence of amendments or clarification on the scope of those laws, private actors can seek governmental authorization through an array of channels, whether they be partnering with law enforcement or seeking authorization to engage in more offensive tactics from the courts in connection with private litigation.

Tags: ,

Posted on November 12, 2025 at 7:01 AM • View Comments

Prompt Injection in AI Browsers

This is why AIs are not ready to be personal assistants:

A new attack called ‘CometJacking’ exploits URL parameters to pass to Perplexity’s Comet AI browser hidden instructions that allow access to sensitive data from connected services, like email and calendar.

In a realistic scenario, no credentials or user interaction are required and a threat actor can leverage the attack by simply exposing a maliciously crafted URL to targeted users.

[…]

CometJacking is a prompt-injection attack where the query string processed by the Comet AI browser contains malicious instructions added using the ‘collection’ parameter of the URL.

LayerX researchers say that the prompt tells the agent to consult its memory and connected services instead of searching the web. As the AI tool is connected to various services, an attacker leveraging the CometJacking method could exfiltrate available data.

In their tests, the connected services and accessible data include Google Calendar invites and Gmail messages and the malicious prompt included instructions to encode the sensitive data in base64 and then exfiltrate them to an external endpoint.

According to the researchers, Comet followed the instructions and delivered the information to an external system controlled by the attacker, evading Perplexity’s checks.

I wrote previously:

Prompt injection isn’t just a minor security problem we need to deal with. It’s a fundamental property of current LLM technology. The systems have no ability to separate trusted commands from untrusted data, and there are an infinite number of prompt injection attacks with no way to block them as a class. We need some new fundamental science of LLMs before we can solve this.

Tags: , , ,

Posted on November 11, 2025 at 7:08 AM • View Comments

New Attacks Against Secure Enclaves

Encryption can protect data at rest and data in transit, but does nothing for data in use. What we have are secure enclaves. I’ve written about this before:

Almost all cloud services have to perform some computation on our data. Even the simplest storage provider has code to copy bytes from an internal storage system and deliver them to the user. End-to-end encryption is sufficient in such a narrow context. But often we want our cloud providers to be able to perform computation on our raw data: search, analysis, AI model training or fine-tuning, and more. Without expensive, esoteric techniques, such as secure multiparty computation protocols or homomorphic encryption techniques that can perform calculations on encrypted data, cloud servers require access to the unencrypted data to do anything useful.

Fortunately, the last few years have seen the advent of general-purpose, hardware-enabled secure computation. This is powered by special functionality on processors known as trusted execution environments (TEEs) or secure enclaves. TEEs decouple who runs the chip (a cloud provider, such as Microsoft Azure) from who secures the chip (a processor vendor, such as Intel) and from who controls the data being used in the computation (the customer or user). A TEE can keep the cloud provider from seeing what is being computed. The results of a computation are sent via a secure tunnel out of the enclave or encrypted and stored. A TEE can also generate a signed attestation that it actually ran the code that the customer wanted to run.

Secure enclaves are critical in our modern cloud-based computing architectures. And, of course, they have vulnerabilities:

The most recent attack, released Tuesday, is known as TEE.fail. It defeats the latest TEE protections from all three chipmakers. The low-cost, low-complexity attack works by placing a small piece of hardware between a single physical memory chip and the motherboard slot it plugs into. It also requires the attacker to compromise the operating system kernel. Once this three-minute attack is completed, Confidential Compute, SEV-SNP, and TDX/SDX can no longer be trusted. Unlike the Battering RAM and Wiretap attacks from last month—which worked only against CPUs using DDR4 memory—TEE.fail works against DDR5, allowing them to work against the latest TEEs.

Yes, these attacks require physical access. But that’s exactly the threat model secure enclaves are supposed to secure against.

Tags: , , ,

Posted on November 10, 2025 at 7:04 AM • View Comments

Faking Receipts with AI

Over the past few decades, it’s become easier and easier to create fake receipts. Decades ago, it required special paper and printers—I remember a company in the UK advertising its services to people trying to cover up their affairs. Then, receipts became computerized, and faking them required some artistic skills to make the page look realistic.

Now, AI can do it all:

Several receipts shown to the FT by expense management platforms demonstrated the realistic nature of the images, which included wrinkles in paper, detailed itemization that matched real-life menus, and signatures.

[…]

The rise in these more realistic copies has led companies to turn to AI to help detect fake receipts, as most are too convincing to be found by human reviewers.

The software works by scanning receipts to check the metadata of the image to discover whether an AI platform created it. However, this can be easily removed by users taking a photo or a screenshot of the picture.

To combat this, it also considers other contextual information by examining details such as repetition in server names and times and broader information about the employee’s trip.

Yet another AI-powered security arms race.

Tags: , ,

Posted on November 7, 2025 at 7:01 AM • View Comments

Rigged Poker Games

The Department of Justice has indicted thirty-one people over the high-tech rigging of high-stakes poker games.

In a typical legitimate poker game, a dealer uses a shuffling machine to shuffle the cards randomly before dealing them to all the players in a particular order. As set forth in the indictment, the rigged games used altered shuffling machines that contained hidden technology allowing the machines to read all the cards in the deck. Because the cards were always dealt in a particular order to the players at the table, the machines could determine which player would have the winning hand. This information was transmitted to an off-site member of the conspiracy, who then transmitted that information via cellphone back to a member of the conspiracy who was playing at the table, referred to as the “Quarterback” or “Driver.” The Quarterback then secretly signaled this information (usually by prearranged signals like touching certain chips or other items on the table) to other co-conspirators playing at the table, who were also participants in the scheme. Collectively, the Quarterback and other players in on the scheme (i.e., the cheating team) used this information to win poker games against unwitting victims, who sometimes lost tens or hundreds of thousands of dollars at a time. The defendants used other cheating technology as well, such as a chip tray analyzer (essentially, a poker chip tray that also secretly read all cards using hidden cameras), an x-ray table that could read cards face down on the table, and special contact lenses or eyeglasses that could read pre-marked cards.

News articles.

Tags: , ,

Posted on November 6, 2025 at 7:02 AM • View Comments

Scientists Need a Positive Vision for AI

For many in the research community, it’s gotten harder to be optimistic about the impacts of artificial intelligence.

As authoritarianism is rising around the world, AI-generated “slop” is overwhelming legitimate media, while AI-generated deepfakes are spreading misinformation and parroting extremist messages. AI is making warfare more precise and deadly amidst intransigent conflicts. AI companies are exploiting people in the global South who work as data labelers, and profiting from content creators worldwide by using their work without license or compensation. The industry is also affecting an already-roiling climate with its enormous energy demands.

Meanwhile, particularly in the United States, public investment in science seems to be redirected and concentrated on AI at the expense of other disciplines. And Big Tech companies are consolidating their control over the AI ecosystem. In these ways and others, AI seems to be making everything worse.

This is not the whole story. We should not resign ourselves to AI being harmful to humanity. None of us should accept this as inevitable, especially those in a position to influence science, government, and society. Scientists and engineers can push AI towards a beneficial path. Here’s how.

The Academy’s View of AI

A Pew study in April found that 56 percent of AI experts (authors and presenters of AI-related conference papers) predict that AI will have positive effects on society. But that optimism doesn’t extend to the scientific community at large. A 2023 survey of 232 scientists by the Center for Science, Technology and Environmental Policy Studies at Arizona State University found more concern than excitement about the use of generative AI in daily life—by nearly a three to one ratio.

We have encountered this sentiment repeatedly. Our careers of diverse applied work have brought us in contact with many research communities: privacy, cybersecurity, physical sciences, drug discovery, public health, public interest technology, and democratic innovation. In all of these fields, we’ve found strong negative sentiment about the impacts of AI. The feeling is so palpable that we’ve often been asked to represent the voice of the AI optimist, even though we spend most of our time writing about the need to reform the structures of AI development.

We understand why these audiences see AI as a destructive force, but this negativity engenders a different concern: that those with the potential to guide the development of AI and steer its influence on society will view it as a lost cause and sit out that process.

Elements of a Positive Vision for AI

Many have argued that turning the tide of climate action requires clearly articulating a path towards positive outcomes. In the same way, while scientists and technologists should anticipate, warn against, and help mitigate the potential harms of AI, they should also highlight the ways the technology can be harnessed for good, galvanizing public action towards those ends.

There are myriad ways to leverage and reshape AI to improve peoples’ lives, distribute rather than concentrate power, and even strengthen democratic processes. Many examples have arisen from the scientific community and deserve to be celebrated.

Some examples: AI is eliminating communication barriers across languages, including under-resourced contexts like marginalized sign languages and indigenous African languages. It is helping policymakers incorporate the viewpoints of many constituents through AI-assisted deliberations and legislative engagement. Large language models can scale individual dialogs to address climate–change skepticism, spreading accurate information at a critical moment. National labs are building AI foundation models to accelerate scientific research. And throughout the fields of medicine and biology, machine learning is solving scientific problems like the prediction of protein structure in aid of drug discovery, which was recognized with a Nobel Prize in 2024.

While each of these applications is nascent and surely imperfect, they all demonstrate that AI can be wielded to advance the public interest. Scientists should embrace, champion, and expand on such efforts.

A Call to Action for Scientists

In our new book, Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship, we describe four key actions for policymakers committed to steering AI toward the public good.

These apply to scientists as well. Researchers should work to reform the AI industry to be more ethical, equitable, and trustworthy. We must collectively develop ethical norms for research that advance and applies AI, and should use and draw attention to AI developers who adhere to those norms.

Second, we should resist harmful uses of AI by documenting the negative applications of AI and casting a light on inappropriate uses.

Third, we should responsibly use AI to make society and peoples’ lives better, exploiting its capabilities to help the communities they serve.

And finally, we must advocate for the renovation of institutions to prepare them for the impacts of AI; universities, professional societies, and democratic organizations are all vulnerable to disruption.

Scientists have a special privilege and responsibility: We are close to the technology itself and therefore well positioned to influence its trajectory. We must work to create an AI-infused world that we want to live in. Technology, as the historian Melvin Kranzberg observed, “is neither good nor bad; nor is it neutral.” Whether the AI we build is detrimental or beneficial to society depends on the choices we make today. But we cannot create a positive future without a vision of what it looks like.

This essay was written with Nathan E. Sanders, and originally appeared in IEEE Spectrum.

Tags: ,

Posted on November 5, 2025 at 7:04 AM • View Comments

← Earlier EntriesLater Entries →

Sidebar photo of Bruce Schneier by Joe MacInnis.