Page 45
Researchers at Google have developed a watermark for LLM-generated text. The basics are pretty obvious: the LLM chooses between tokens partly based on a cryptographic key, and someone with knowledge of the key can detect those choices. What makes this hard is (1) how much text is required for the watermark to work, and (2) how robust the watermark is to post-generation editing. Google’s version looks pretty good: it’s detectable in text as small as 200 tokens.
An advocacy groups is filing a Fourth Amendment challenge against automatic license plate readers.
“The City of Norfolk, Virginia, has installed a network of cameras that make it functionally impossible for people to drive anywhere without having their movements tracked, photographed, and stored in an AI-assisted database that enables the warrantless surveillance of their every move. This civil rights lawsuit seeks to end this dragnet surveillance program,” the lawsuit notes. “In Norfolk, no one can escape the government’s 172 unblinking eyes,” it continues, referring to the 172 Flock cameras currently operational in Norfolk. The Fourth Amendment protects against unreasonable searches and seizures and has been ruled in many cases to protect against warrantless government surveillance, and the lawsuit specifically says Norfolk’s installation violates that.”
The headline is pretty scary: “China’s Quantum Computer Scientists Crack Military-Grade Encryption.”
No, it’s not true.
This debunking saved me the trouble of writing one. It all seems to have come from this news article, which wasn’t bad but was taken wildly out of proportion.
Cryptography is safe, and will be for a long time
EDITED TO ADD (11/3): Really good explainer from Dan Goodin.
The Wall Street Journal is reporting that the CEO of a still unnamed company has been indicted for creating a fake auditing company to falsify security certifications in order to win government business.
EDITED TO ADD (11/14): More info.
The men’s world conkers champion is accused of cheating with a steel chestnut.
The Washington Post has a long and detailed story about the operation that’s well worth reading (alternate version here).
The sales pitch came from a marketing official trusted by Hezbollah with links to Apollo. The marketing official, a woman whose identity and nationality officials declined to reveal, was a former Middle East sales representative for the Taiwanese firm who had established her own company and acquired a license to sell a line of pagers that bore the Apollo brand. Sometime in 2023, she offered Hezbollah a deal on one of the products her firm sold: the rugged and reliable AR924.
“She was the one in touch with Hezbollah, and explained to them why the bigger pager with the larger battery was better than the original model,” said an Israeli official briefed on details of the operation. One of the main selling points about the AR924 was that it was “possible to charge with a cable. And the batteries were longer lasting,” the official said.
As it turned out, the actual production of the devices was outsourced and the marketing official had no knowledge of the operation and was unaware that the pagers were physically assembled in Israel under Mossad oversight, officials said. Mossad’s pagers, each weighing less than three ounces, included a unique feature: a battery pack that concealed a tiny amount of a powerful explosive, according to the officials familiar with the plot.
In a feat of engineering, the bomb component was so carefully hidden as to be virtually undetectable, even if the device was taken apart, the officials said. Israeli officials believe that Hezbollah did disassemble some of the pagers and may have even X-rayed them.
Also invisible was Mossad’s remote access to the devices. An electronic signal from the intelligence service could trigger the explosion of thousands of the devices at once. But, to ensure maximum damage, the blast could also be triggered by a special two-step procedure required for viewing secure messages that had been encrypted.
“You had to push two buttons to read the message,” an official said. In practice, that meant using both hands.
Also read Bunnie Huang’s essay on what it means to live in a world where people can turn IoT devices into bombs. His conclusion:
Not all things that could exist should exist, and some ideas are better left unimplemented. Technology alone has no ethics: the difference between a patch and an exploit is the method in which a technology is disclosed. Exploding batteries have probably been conceived of and tested by spy agencies around the world, but never deployed en masse because while it may achieve a tactical win, it is too easy for weaker adversaries to copy the idea and justify its re-deployment in an asymmetric and devastating retaliation.
However, now that I’ve seen it executed, I am left with the terrifying realization that not only is it feasible, it’s relatively easy for any modestly-funded entity to implement. Not just our allies can do this—a wide cast of adversaries have this capability in their reach, from nation-states to cartels and gangs, to shady copycat battery factories just looking for a big payday (if chemical suppliers can moonlight in illicit drugs, what stops battery factories from dealing in bespoke munitions?). Bottom line is: we should approach the public policy debate around this assuming that someday, we could be victims of exploding batteries, too. Turning everyday objects into fragmentation grenades should be a crime, as it blurs the line between civilian and military technologies.
I fear that if we do not universally and swiftly condemn the practice of turning everyday gadgets into bombs, we risk legitimizing a military technology that can literally bring the front line of every conflict into your pocket, purse or home.
Perfectl in an impressive piece of malware:
The malware has been circulating since at least 2021. It gets installed by exploiting more than 20,000 common misconfigurations, a capability that may make millions of machines connected to the Internet potential targets, researchers from Aqua Security said. It can also exploit CVE-2023-33246, a vulnerability with a severity rating of 10 out of 10 that was patched last year in Apache RocketMQ, a messaging and streaming platform that’s found on many Linux machines.
The researchers are calling the malware Perfctl, the name of a malicious component that surreptitiously mines cryptocurrency. The unknown developers of the malware gave the process a name that combines the perf Linux monitoring tool and ctl, an abbreviation commonly used with command line tools. A signature characteristic of Perfctl is its use of process and file names that are identical or similar to those commonly found in Linux environments. The naming convention is one of the many ways the malware attempts to escape notice of infected users.
Perfctl further cloaks itself using a host of other tricks. One is that it installs many of its components as rootkits, a special class of malware that hides its presence from the operating system and administrative tools. Other stealth mechanisms include:
- Stopping activities that are easy to detect when a new user logs in
- Using a Unix socket over TOR for external communications
- Deleting its installation binary after execution and running as a background service thereafter
- Manipulating the Linux process pcap_loop through a technique known as hooking to prevent admin tools from recording the malicious traffic
- Suppressing mesg errors to avoid any visible warnings during execution.
The malware is designed to ensure persistence, meaning the ability to remain on the infected machine after reboots or attempts to delete core components. Two such techniques are (1) modifying the ~/.profile script, which sets up the environment during user login so the malware loads ahead of legitimate workloads expected to run on the server and (2) copying itself from memory to multiple disk locations. The hooking of pcap_loop can also provide persistence by allowing malicious activities to continue even after primary payloads are detected and removed.
Besides using the machine resources to mine cryptocurrency, Perfctl also turns the machine into a profit-making proxy that paying customers use to relay their Internet traffic. Aqua Security researchers have also observed the malware serving as a backdoor to install other families of malware.
Something this complex and impressive implies that a government is behind this. North Korea is the government we know that hacks cryptocurrency in order to fund its operations. But this feels too complex for that. I have no idea how to attribute this.
Fishermen in Tamil Nadu are reporting smaller catches of squid.
In July, I wrote about my new book project on AI and democracy, to be published by MIT Press in fall 2025. My co-author and collaborator Nathan Sanders and I are hard at work writing.
At this point, we would like feedback on titles. Here are four possibilities:
What we want out of the title is that it convey (1) that it is a book about AI, (2) that it is a book about democracy writ large (and not just deepfakes), and (3) that it is largely optimistic.
What do you like? Feel free to do some mixing and matching: swapping “Will Transform” for “Will Improve” for “Can Transform” for “Can Improve,” for example. Or “Democracy” for “the Republic.” Remember, the goal here is for a title that will make a potential reader pick the book up off a shelf, or read the blurb text on a webpage. It needs to be something that will catch the reader’s attention. (Other title ideas are here).
Also, FYI, this is the current table of contents:
Introduction
1. Introduction: How AI will Change Democracy
2. Core AI Capabilities
3. Democracy as an Information SystemPart I: AI-Assisted Politics
4. Background: Making Mistakes
5. Talking to Voters
6. Conducting Polls
7. Organizing a Political Campaign
8. Fundraising for Politics
9. Being a PoliticianPart II: AI-Assisted Legislators
10. Background: Explaining Itself
11. Background: Who’s to Blame?
12. Listening to Constituents
13. Writing Laws
14. Writing More Complex Laws
15. Writing Laws that Empower Machines
16. Negotiating LegislationPart III: The AI-Assisted Administration
17. Background: Exhibiting Values and Bias
18. Background: Augmenting Versus Replacing People
19. Serving People
20. Operating Government
21. Enforcing RegulationsPart IV: The AI-Assisted Court
22. Background: Being Fair
23. Background: Getting Hacked
24. Acting as a Lawyer
25. Arbitrating Disputes
26. Enforcing the Law
27. Reshaping Legislative Intent
28. Being a JudgePart V: AI-Assisted Citizens
29. Background: AI and Power
30. Background: AI and Trust
31. Explaining the News
32. Watching the Government
33. Moderating, Facilitating, and Building Consensus
34. Acting as Your Personal Advocate
35. Acting as Your Personal Political ProxyPart VI: Ensuring That AI Benefits Democracy
36. Why AI is Not Yet Good for Democracy
37. How to Ensure AI is Good for Democracy
38. What We Need to Do Now
39. Conclusion
Everything is subject to change, of course. The manuscript isn’t due to the publisher until the end of March, and who knows what AI developments will happen between now and then.
EDITED: The title under consideration is “Rewiring the Republic,” and not “Rewiring Democracy.” Although, I suppose, both are really under consideration.
Sidebar photo of Bruce Schneier by Joe MacInnis.