---
title: "Oct 29: ICE surveillance of political activists, and more"
show: "RobotPaper's daily HN podcast"
canonical: "https://podcasts.robotpaper.ai/hn/ep/2026-10-29/"
date: "2026-10-29T03:25:46+00:00"
duration: "28:20"
duration_seconds: 1700
audio: "https://podcasts.robotpaper.ai/hn/ep/2026-10-29/2026-10-29-hn.mp3"
chapters: "https://podcasts.robotpaper.ai/hn/ep/2026-10-29/chapters.json"
transcript: "https://podcasts.robotpaper.ai/hn/ep/2026-10-29.txt"
feed: "https://podcasts.robotpaper.ai/hn/feed.xml"
language: "en"
author: "Roy Osherove"
ai_generated_voices: true
---

# Oct 29: ICE surveillance of political activists, and more

RobotPaper's daily HN podcast · Oct 29, 2026 · 28 min

- Listen: [MP3, 27 MB](https://podcasts.robotpaper.ai/hn/ep/2026-10-29/2026-10-29-hn.mp3)
- Subscribe via RSS: https://podcasts.robotpaper.ai/hn/feed.xml (paste into any podcast app)
- Web page: https://podcasts.robotpaper.ai/hn/ep/2026-10-29/
- Chapters (JSON): https://podcasts.robotpaper.ai/hn/ep/2026-10-29/chapters.json
- Transcript (text): https://podcasts.robotpaper.ai/hn/ep/2026-10-29.txt
- Show: [RobotPaper's daily HN podcast](https://podcasts.robotpaper.ai/hn/) (RSS: https://podcasts.robotpaper.ai/hn/feed.xml)

## Chapters

- `00:00:00` Intro
- `00:00:38` ICE surveillance
- `00:03:14` MacBook touch display
- `00:05:38` Gemini Carbon
- `00:08:05` Humanist AI exemption
- `00:10:22` Wealth to scale
- `00:12:23` Apple certificate gap
- `00:14:21` Claude drug research
- `00:16:30` US Code history
- `00:18:18` UEFI boot flaw
- `00:20:15` Toledo train collision
- `00:22:12` The Grid PCB factory
- `00:22:55` Discord web search
- `00:23:31` Offline emergency AI
- `00:24:14` Browser fingerprinting
- `00:24:56` Tailscale SSH flaws
- `00:25:42` Emacs anniversary
- `00:26:25` Serbian election
- `00:27:01` AI data worker map
- `00:27:46` Outro

## Show notes

- Evidence of a spying operation by ICE against left-wing social movements
- MacBook Pro with PROLED XDR Display and Touch
- Gemini 4 Carbon
- Kentucky man becomes the first humanist granted religious exemption from AI
- Wealth Inequality in America drawn to scale
- Apple's git still trusts the compromised eMudhra certificates
- Early progress on accelerating drug development with Claude
- Git for the US Code
- Code-execution flaw in UEFI firmware threatens most servers on planet Earth
- Two high-speed trains collide near Toledo, Spain
- The Grid, a factory-scale robot in Massachusetts that builds PCBs on demand
- Discord will soon let anyone find public server messages through Google
- Survival LLM: AI assistant and emergency library that runs on your own computer
- Why is this fingerprinting script creating bioweapons?
- Tailscale SSH: Investigation and remediation update
- Emacs Turns 50
- Serbia's Vucic concedes election defeat in historic win for student movement
- Brown student maps the hidden global labor powering AI

The voices you're hearing are AI-generated; the stories come from public sources linked in the show notes.\
Music: Eleven Music.\
An independent project; not affiliated with or endorsed by Amazon Web Services, Y Combinator, Hacker News or any vendor or news outlet mentioned. Opinions are the hosts' own.\
Produced by Roy Osherove.

## Transcript

\[Linus\] Good morning, it's Thursday, October twenty-ninth, and this is RobotPaper's daily HN podcast. I'm Linus. The voices you're hearing are AI-generated; the stories come from news dot Y Combinator dot com.

\[Dennis\] I'm Dennis, and today, surveillance records raise questions about ICE's investigation of political activists.

\[Ada\] And I'm Ada. Apple adds touch to MacBook Pro; Google announces Gemini four Carbon.

\[Grace\] I'm Grace, and as always, this podcast is generated by robotpaper dot A I, by Roy Osherove. Let's go.

\[Dennis\] First up, from Wired: evidence of an ICE spying operation against left-wing social movements. Records obtained by defense lawyers in a Minnesota conspiracy case show how agents tracked activists who oppose immigration enforcement. The records include reports, transcripts, photos, database searches and messages from Signal groups that agents had joined. Wired says around one hundred twenty Homeland Security Investigations agents worked on the case, including eleven undercover agents. The records name Project Whipple Shield and Operation Puppet Master, and say the work was authorized under National Security Presidential Memorandum Seven. Agents used Reaper drones and powers normally linked to terrorism and money laundering cases. Undercover staff went to community meetings, marches and training sessions. Agents also tracked activist meetings outside Minnesota. Prosecutors say the defendants conspired to block immigration enforcement, identify and stalk federal agents, and assault them. Defense lawyers say the charges do not match the alleged acts and that the case is part of a wider campaign to spy on political opponents. The report links the case to Operation Keyhole, a program set up in two thousand eight to fund undercover work. It seems the story is drawing interest because powers built for serious crime cases are being used to track political groups, while the full scope of the program is still unclear.

\[Ken\] And the thread is deeply concerned about surveillance, with a dispute over the line between crime probes and tracking dissent. Multiplegeorges focuses on the way cloud records, payment data and commercial databases were combined, and asks how agents got SSH session logs. John\_strinlai points to case records that list network data from Fly, including new session entries and outbound connections. The concern is how separate records can be used together to identify someone. Brightball asks whether the probe was a response to organized anti-ICE activity, saying the headline leaves out that context. Someone1234 asks why ICE should be probing that activity at all. Elsewhere, nomilk asks why enforcing immigration rules draws such opposition. Jnovek replies that the objection is to force and intimidation, rather than immigration law itself. The exchanges turn on whether an enforcement goal justifies the methods used, and how much data service providers and brokers make available to agents.

\[Linus\] From Apple: MacBook Pro with PROLED XDR Display and Touch. The new design pairs a tandem OLED touchscreen with a lighter case and a new thermal system. Apple lists fourteen-inch and sixteen-inch models weighing three and four pounds. Touch gestures let people scroll, zoom, draw and move objects, with the touch sensor built into the panel. Apple says the screen has up to two thousand nits of peak brightness for both standard and high dynamic range content, and up to one thousand nits of sustained full-screen brightness. The panel covers the P three and Adobe RGB color gamuts. Other changes include Dynamic Island, slimmer display borders, an updated six-speaker sound system and a twelve-megapixel Center Stage camera. The page lists M five Pro and M five Max chips for these models, with up to one hundred twenty-eight gigabytes of unified memory on M five Max. Its screen comparison sets these models apart from the mini-LED MacBook Pro. Ports include three Thunderbolt five sockets, HDMI, MagSafe and an SDXC card slot. It seems the significance goes beyond screen quality, since direct touch adds a new way to use the Mac's desktop interface.

\[Grace\] The comments are split between praise for the hardware and concern about what touch could do to macOS. CountSessine argues that touch will create pressure to enlarge controls and change the interface for everyone, including those who never touch the screen. Grombobulous disagrees, saying touch on Windows and Linux laptops is an option that can work alongside desktop controls. Other comments describe how people use their devices. Rubicon33 says their spouse wants a keyboard, desktop apps and touch together, and moves between typing and using the screen directly. Joshstrange prefers the accuracy of a mouse and questions the value for a laptop that spends most of its time docked. Meanwhile, tristanj praises the lower weight, while reenorap says the price of an equivalent high-memory model has risen a lot since their purchase. Raydev notes that this price comparison also includes a better screen with touch. The debate covers both the future of the interface and the value of the changes to the hardware.

\[Ada\] Next, from Google's blog: Gemini four Carbon. Google has announced a model aimed at cybersecurity and long software tasks, with access at first limited to selected defenders and trusted partners. The company says Carbon improves on Gemini four Argon, announced two weeks earlier. Google reports eighty-eight point two percent on SEC-bench Pro for finding flaws, fifty-four point eight percent on ExploitGym, and seventy-one percent on CWE-bench for patching. Those external tests were run without guardrails. Google says its own teams have seen the model work for days on complex engineering tasks, including finding and fixing bugs across large codebases. It also reports fewer claims without evidence and fewer high-severity misalignment alerts in internal use. Access will grow through the Fairwind Program as Google gathers feedback and adjusts safeguards. The company says it is taking part in the United States government's voluntary process for testing models before release. The stated launch price matches Argon at two dollars per million input tokens and ten dollars per million output tokens. It seems the next question is how the reported gains will translate into tools that work well once more people have access.

\[Dennis\] In the comments, complaints about access and product names often draw more focus than the test results. The commenter eight two seven a describes trouble finding the latest model and knowing which plan is needed. Snoman says the mix of version numbers and product tiers makes the model selector hard to use. RajuChacha108 proposes fewer, stable tiers, while yipinwong says users can settle on names without a rule for the whole industry. Solarpunk\_enthu questions the value of announcing gains before a model is widely available, since rivals may advance before release. Hiddencost replies that Argon became available soon after its announcement and argues for giving this rollout time. Reports of Argon's performance also differ. Nater5000 describes good value and easy access through Cursor at work, while seizethecheese reports poor results in a local coding-agent comparison. The thread puts access, ease of use and results on real work alongside the company's reported scores.

\[Ken\] From Religion News Service: a Kentucky man becomes the first humanist granted a religious exemption from AI, according to the American Humanist Association. The University of Kentucky has agreed that research administrator David Durman does not have to use AI to do his work. Durman says handing off thought and decisions conflicts with his humanist beliefs about moral duty and human flourishing. With help from the association, he asked for an accommodation under Title Seven. The university granted it in September. Durman helps researchers with grant applications and says he can do the job without AI or affecting his co-workers. Association lawyer Alexandria Huiskens says humanism is a comprehensive way of life and that its followers should have room to act on those beliefs at work. The article also quotes law professor Douglas Laycock, who says these requests depend on a sincere ethical commitment like a religious one. It seems the case draws interest because it brings freedom of conscience into choices about workplace tools. The university granted this request, rather than a court ruling that set a general right to refuse AI.

\[Linus\] And the thread divides between respect for personal beliefs and concern about keeping pace as work changes. Aurornis compares refusing AI with earlier refusals to use email, arguing that careers require people to adapt. Mattw2121 asks what happens if future tasks cannot be done by their deadlines without AI. Bunderbunder replies that deadlines should change when they conflict with understanding a problem and taking responsibility for a decision. Another exchange concerns what this case establishes. Cvoss stresses that the employer granted the request of its own accord and that no court decided this dispute. Humorist2290 replies to the broader claim about humanism by pointing to the court case cited in the article that recognized it as a religion. Joshstrange raises a separate objection, asking why handing thought to AI differs from using earlier computing tools. The thread thus links the needs of a job with debates about conscience, responsibility and what sets these tools apart.

\[Grace\] Next up, from Dream Again USA: Wealth Inequality in America, Drawn to Scale. A chart lets users scroll through American household wealth, on one scale from the poorest homes to the richest. Each pixel stands for five thousand dollars, and each column stands for one percent of households. The site shows stacks of hundred dollar bills: about nine inches for the median household, against a hundred and forty-five feet for the average household in the top one percent. The data mixes sources, rather than using one uniform series. Most columns use the twenty twenty-three Survey of Consumer Finances, with values raised to twenty twenty-six dollars. The top one percent uses a mean from the Federal Reserve's Distributional Financial Accounts for the first quarter of this year. The poorest eight percent still show their reported debt. A version to print is on the site. The appeal seems to be a sense of scale, and a prompt to ask what wealth gives its owners.

\[Ada\] And the thread is split between calls to tax large fortunes and doubts about how to treat wealth held in assets. TrackerFF said stakes in firms make up much of that wealth, so a yearly wealth tax could force owners to pay out dividends, sell shares or borrow. Etheryte replied that a tax is meant to reduce assets, so that alone is not a case against it. Solatic drew on their own startup shares, which were hard to value or sell, to ask if such wealth brings real financial security. The same comment asked why so much company wealth grows outside public markets. Bryanlarsen stressed that wealth also buys power in politics and the economy, even if the owner does not spend it. Arjie said higher living standards matter more than equality. Modo\_mario replied that large holdings can affect access to basic resources, such as farmland and timber. The debate goes well past the chart to cash, power and the aims of tax.

\[Dennis\] From We Have A Problem: Apple's Git Still Trusts the Compromised eMudhra Certificates. Tools shipped with macOS still trusted the bad certificates after Apple had protected Safari and apps that use its own network framework. The October twenty-eighth report says Apple was told of the breach on October seventeenth. It says macOS twenty-seven point two left the flaw in place and undid a fix some users had made by hand. Git, curl and Python are among the tools at risk. An attacker on their network path could pose as an HTTPS service, change code sent through Git, or steal login details. The scope is worth spelling out: the flaw is in the checks on certificates, and an attack also needs control of the network path. The site offers checks and a demo, and says users must check again after each operating system update until Apple ships a fix. It also flags risk to macOS build machines. This seems key because a safe browser does not mean each tool has the same checks.

\[Ken\] Over in the comments, anger at Apple's response meets debate over how easily an attack could work. Chrisweekly stressed that this goes beyond Git to a range of system tools. Stingraycharles said it makes sense to trust the operating system, so the fault lies with macOS. Woodruffw, who said they help maintain Homebrew, first pointed to its own certificate bundle, then added a correction: its Git was still at risk and a fix was in progress. That shows why a tool from a new source is not always safe. Paaloeye said the threat was overstated since an attack needs further access. Treesknees cited client isolation and network defenses that make local attacks harder. Dspillett urged readers not to dismiss the flaw, since a bad signing certificate could help as part of a larger attack.

\[Linus\] Next, from Anthropic: Early Progress on Accelerating Drug Development with Claude. Anthropic has shared early lab and paperwork results from work across several stages of drug research. The firm says it has worked with six groups and begun drug programs, including rare childhood forms of muscular dystrophy. At Generate Biomedicines, scientists used Claude Code to suggest shorter genetic instructions for part of CAR T-cell therapy. Anthropic says one design was about seventy percent shorter, let cells make more receptors and worked better in lab tests. At Enveda, a document checker assessed more than a thousand pages. It caught eleven issues that human reviewers had found, and flagged seven they had missed. Anthropic says human review time fell from more than twenty-five hours to less than three. These are early tests of different tasks, not proof that a finished drug has passed clinical trials. The value seems to lie in helping with the work between a sound idea for a drug and a treatment for patients.

\[Grace\] The thread mixes hope for new drugs with doubts about which tasks these tools can speed up. Orthogonal\_cube welcomed the prospect of cheaper research and drugs for rare diseases, but found Anthropic's pitch self-serving. Gumby, a self-described former drug developer, argued that coming up with candidates is already the easy part. Bahmboo replied that the article also covers making drugs, filing with regulators and running trials, as well as molecule design. Timr said the share of costs for each task matters, and clinical trials still shape the economics. Jrowen saw a role helping each expert study data, plan experiments and manage information. Enraged\_camel stressed how long it takes to test results through lab work, clinical trials and regulatory steps. That leaves a gap between hopes for help and claims about how fast new drugs can reach patients.

\[Ada\] Over on the Lexifina blog: Git for the US Code. A legal history project links changes in federal law to the laws that made them and the dates they took effect. The team applied amendments across nine titles, using release points since twenty seventeen. Of three hundred and sixty-five changes, three hundred and sixty-one matched the next official version exactly. The other four have labels stating how far they could be checked. Users can pick a date, see the text in force, find which Act wrote a phrase, and track changes to a section. There is a key limit: the tool does not yet show how a later law can change the legal treatment of earlier events. If an effective-date note is hard to parse, it can use the law's enactment date. Each change links to the original law to check. This seems useful because a search or language model can cite the wrong version of a law, even if the words look sound.

\[Dennis\] And the comments welcome easier access to history, but split on whether version control makes politics more open. Kfogel cited an earlier plan to track who put words into bills. P-e-w said the press still needs to find out who wrote them and explain the context, since a commit log cannot offer oversight. Codingdave, who described fifteen years building document systems for small government bodies, backed version control but asked if Git was the right tool. They said lawmaking workflows already track amendments and publish new text. Jcranmer added that to grasp a rule's meaning, users may need the debate and intent behind it. A history of the text is one task, and explaining who shaped the law and what it means is another.

\[Ken\] From Ars Technica: A Code-Execution Flaw in UEFI Firmware Threatens Most Servers on Planet Earth. A memory corruption bug affects a widely used firmware codebase during network boot. The bug lies in how TianoCore EDK Two's PXE boot driver handles ICMP error messages. A message that is too large can overwrite a fixed buffer and may run code before the operating system's defenses are in place. An attacker must reach the right network interface during boot, and running code also requires knowledge of the target firmware. The payload itself does not last after a reboot, but it could help install further harmful code that does. The report says a public patch preceded a joint disclosure planned for January, which brought forward an advisory. The concern seems to be the code's high level of access and the work needed to send firmware fixes through many vendors. The broad headline does not mean every machine can be reached or uses the boot path at risk.

\[Linus\] The thread is split between alarm at firmware complexity and doubts about the scope implied by the headline. FrasiertheLion described a much smaller firmware design for specific confidential virtual machines, saying a narrow set of needs can reduce attack surface. Dspillett warned that small parts can still add up to more dependencies. VoidUpdate questioned the short time window during boot. Nicce replied that attackers could keep sending packets and wait for a restart. Kikdij noted that the operating system's firewall cannot protect it before it starts. Tptacek said shared management networks at bare-metal hosts could make access easier than readers think. The debate turns on how machines boot, who can reach their networks and what powers an attack could gain.

\[Grace\] Finally, from Euronews: Two High-Speed Trains Collide Near Toledo, Spain. A train from Madrid towards Seville hit the rear of another train that had stopped on the same track. It was between Urda and Los Yébenes in Toledo province. Euronews says regional government spokeswoman Esther Padilla confirmed ten deaths and warned that the toll was likely to rise. A Renfe train carried three hundred and ninety-seven passengers, while an Iryo service had a hundred and forty-seven people aboard. Rescue teams focused on the cars with the worst damage, and those hurt were taken to nearby hospitals. A military emergency unit helped, and services between Madrid, Ciudad Real and Andalusia were stopped. The report says the cause remains unknown. This tragedy seems likely to bring more scrutiny of rail safety, after another fatal crash in Spain earlier this year.

\[Ada\] And the thread is distressed, with a split over blame and how trains should run if automated protection fails. Purpleidea blamed the crash on broader incompetence. Paimapi pushed back, saying staffing, working conditions and managers' choices need scrutiny, not just the conduct of each worker. Renehsz stressed the need for clear fallback rules when train protection or track detection systems fail. That includes moving with care where crews cannot be sure the track is clear. These concern safety rules, while the Euronews report leaves the cause of this crash unresolved. SilverFork proposed public access to train location data as an extra way to monitor them. Fmpfmp asked if GPS could reliably tell apart nearby parallel tracks. Others compared rail with road and air travel, but the core concern was how a railway should preserve safe separation if its normal protections become unavailable.

\[Dennis\] A few more that did not make the cut today. From CircuitHub: The Grid, a factory-scale robot in Massachusetts that builds PCBs on demand. The system automates circuit board assembly for varied designs and small production runs on demand. Levitating movers carry boards and components between machines for printing solder paste, placing parts, soldering and inspection. CircuitHub says models, AI agents and optimizers schedule the work. Commenters welcomed domestic manufacturing capacity but questioned pricing, board restrictions and whether the speed claims reflect their experience with Chinese suppliers.

\[Ken\] Next, from the Discord blog: public server messages will become discoverable through web search. Discord is introducing web pages for its Discoverable Servers, with some public messages appearing later. Server pages begin rolling out November sixteenth, while messages may appear from mid-December during a gradual rollout. Administrators can opt individual channels or all channels out. Discoverable Servers have at least a thousand members. Commenters welcomed access to knowledge previously difficult to search, while others raised privacy concerns and questioned why smaller communities would be excluded from the change.

\[Linus\] From Brangus: Survival LLM, an AI assistant and emergency library that runs on your own computer. This free, open source application combines local models with reference material for use without internet access. Its library includes medical references, federal emergency guidance, repair manuals and about nine hundred thirty thousand Wikipedia articles. Answers link to the source passages, and model selection depends on available memory. Commenters were divided between seeing AI as useful search assistance and worrying that plausible mistakes, demanding hardware requirements and unreliable emergency advice outweigh that convenience.

\[Grace\] Over on Nullpt: Why is this fingerprinting script creating bioweapons? The post investigates browser fingerprinting scripts through the graphics they generate. The author crawled ten thousand websites, intercepting browser calls associated with fingerprinting, and collected the results in a gallery. Canvas fingerprinting uses differences in how browsers, operating systems and graphics hardware render images to distinguish browser environments. Commenters praised the investigation and discussed techniques that obstruct automated analysis. Their reactions also reflected disagreement over whether AI safeguards protect users or prevent legitimate technical work from being completed.

\[Ada\] From Tailscale: Tailscale SSH investigation and remediation update. The company has published four detailed security bulletins describing vulnerabilities, affected configurations and recommended customer actions. Most identified issues concern privilege escalation, potentially allowing someone already authorized to connect through Tailscale SSH to exceed the permissions of their local user. Exploitation depends on access permissions, configuration and other conditions. Customers are directed to the company’s applicable update guidance. Commenters debated the impact, with some favoring simpler OpenSSH or WireGuard configurations and others defending Tailscale's value as one layer of security.

\[Dennis\] From MIT Technology Review: Emacs turns fifty. The enduring editor began running on video terminals at the MIT Artificial Intelligence Laboratory in nineteen seventy-six. Its concepts survived rewrites for many operating systems, and its early software-sharing community helped give rise to the free software movement and GNU/Linux. The article traces its roots to the earlier TECO editor and its programmable editing commands. Commenters celebrated decades of continued daily use and active development, while others discussed keeping configurations manageable and whether extensive customization helps or distracts from everyday work.

\[Ken\] From Reuters: Serbia's Vucic concedes election defeat in a historic win for the student movement. A student-led protest movement has won the election, ending Vucic's long grip on national political power. He called the snap vote after nearly two years of demonstrations against perceived corruption and growing authoritarianism. Those protests followed the deaths of sixteen people in a station canopy collapse in Novi Sad in twenty twenty-four. Commenters largely celebrated the result, while skeptics questioned how much political change would follow and debated descriptions of his government's record.

\[Linus\] From Brown University: a student maps the hidden global labor powering AI. Undergraduate Cunyan Ma has created an interactive map connecting technology companies and intermediaries with data workers located around the world. Freely available online, it combines existing data worker databases, original data collection and reporting on working conditions. The project aims to make the human labor behind AI more visible. Commenters sharply disagreed over the framing of AI colonialism, with some emphasizing worker isolation and unequal power, while others argued that outsourced jobs can improve local employment opportunities.

\[Ken\] That's RobotPaper's daily HN podcast for Thursday, October twenty-ninth. Today's coverage included surveillance records, Apple's touch display, and Google's restricted model rollout, alongside research, security and public infrastructure. This podcast is generated by robotpaper dot A I, by Roy Osherove. For more podcasts from RobotPaper, head over to podcasts dot robotpaper dot A I.

---

The voices you're hearing are AI-generated; the stories come from public sources linked in the show notes. An independent project; not affiliated with or endorsed by Amazon Web Services, Y Combinator, Hacker News or any vendor or news outlet mentioned. Opinions are the hosts' own. Produced by Roy Osherove. Privacy: We count visits to these pages with Cloudflare Web Analytics (cookieless). Google Analytics (which sets cookies): visitors in the EU/EEA, the UK and Switzerland are asked first and it runs only if they accept (we go by the country Cloudflare reports for the connection; if it is unknown, we ask). Elsewhere it loads by default. Anyone can turn it off with the Privacy settings button at the bottom of every page; the choice is kept for a year in this browser (local storage "analytics-consent"). We count feed fetches and downloads in aggregate; IP addresses are hashed and never stored. If you send feedback, we keep what you write (and your email if you give it) to read and reply. © 2026 Roy Osherove.
