Weekly Surveillance Roundup, October 26 to November 1, 2026

The Throughline: Nobody Can Produce the Paperwork

Read the week's stories one at a time and you see a pile of unrelated news. Read them as one beat and the shape is obvious. The power to collect, to watch, to scan, to store is spreading faster than the power to document what happened. A foreign intelligence court criticized intelligence agencies for using filter tools to search data for Americans without recording the activity as queries, and that ruling has sat in a drawer past its statutory deadline ^[1][3]. GCHQ lost an unknown number of removable storage devices and cannot produce a number ^[5][6]. The Medicare AI that denies coverage sits behind more than 800 pages of records that EFF had to sue for, and the pages do not show how much human review any denial actually got ^[8]. Quad9 would rather stop serving a whole country than hand over a list it cannot guarantee is complete ^[16]. The UK government wants a working age-verification system for a social media ban, and Ofcom's shortlist is four ways to turn your face, your bank, your identity document, or your photos into a credential ^[9]. Meanwhile the largest multi-tenant intrusion in this week's reporting, the one against more than 2,250 cloud tenants, only has a shape because Microsoft decided to publish it ^[10][11].

The second throughline is that the tools are getting ahead of the rules. The ConnectWise compromise handed attackers a high-privilege application that let them operate across customer tenants, and Microsoft's own framing is that successful authentication "does not necessarily indicate that customer data was accessed or exfiltrated" ^[10]. LiteLLM's master key sat in a public GitHub repository from May 27 until October 10, and a poisoned release reached roughly 49,000 downloads before PyPI quarantined it ^[13]. The Miasma worm hid its payload in blobs large enough to evade GitHub's search and API paths, and its hooks targeted Claude, Gemini, Cursor, and VS Code ^[14]. Every one of those systems is an automation layer that runs faster than the humans who are supposed to supervise it, and every one of them leaves an audit trail that the people affected never get to read.

The third throughline is money and consent, and it is the part that touches you. GM's drivers had their location, speed, and seatbelt habits collected and sold to insurers who used it to set premiums, and more than 370,000 of them were told through the OnStar app that they were not enrolled ^[7]. Meta's glasses captured intimate imagery of people who never agreed to be recorded, and now two senators want answers ^[15]. The pattern is the same one the FISA ruling describes, just smaller and slower: a system collects first, tells you later if it tells you at all, and asks you to fix the record yourself when the record is wrong.

The FISA Court Ruling Nobody Will Publish

On October 26, Sen. Ron Wyden and Sen. Adam Schiff wrote to Director of National Intelligence Jay Clayton and Attorney General Todd Blanche about a court decision that has never been published. Tim Starks at CyberScoop reported that the decision at issue is a March 17, 2026 opinion from the Foreign Intelligence Surveillance Court that criticized government-created tools for processing intercepted messages, and CDT reported that Wyden and Schiff separately demanded release and the identities of the officials delaying it ^[1][2]. Wyden said the court identified surveillance problems "directly related to Americans' constitutional rights," and an unnamed Senate Intelligence Committee member called the findings "major compliance problems" ^[2]. Jake Laperruque at Just Security reported on October 29 that Wyden and Rep. Thomas Massie have both read the opinion. Massie said it "raises serious concerns about FBI implementation of FISA 702." Wyden said the court "found serious violations of Americans' constitutional rights" ^[3].

What the opinion found is the part that makes it worth hiding. Per Laperruque's account, the FBI, CIA, and NSA accessed Americans' communications without following the procedures required for running queries, which meant those searches escaped "any audit or oversight review" and were left out of the annual count of how many queries agencies run ^[3]. A search with no approval, no written justification, no audit trail, and no line in the totals is a search nobody outside the agency can see happening. That sits badly with the number the administration is leading with: FBI Director Kash Patel told a Senate hearing that "For the first time in the FBI's history, our queries, per the FISA Court itself, are over 99 percent compliant," and his written testimony said the bureau put in place "a continuous double audit" of all Section 702 queries ^[3]. A compliance rate that counts the queries on the books does not count the queries the court says were run off the books ^[3].

The statute gives the government 180 days and a fallback. CDT explained in its October 26 letter that under 50 U.S.C. § 1872 the DNI and the attorney general must review significant FISA Court opinions for declassification and publish them with only the redactions needed to protect legitimate sources and methods; for this opinion that window closed on September 13 ^[2]. If no portion can be released, the law still requires an official statement of the opinion's significant legal conclusions ^[2]. The statute "gives no option for indefinite silence," the letter says, and warns that "the executive branch simply cannot be relied on to police itself" ^[2]. More than 20 organizations signed, including the ACLU, EFF, EPIC, the Brennan Center for Justice, Common Cause, Demand Progress, Fight for the Future, and Muslim Advocates ^[2]. The delay is not new either. Cotton and Warner asked for release within 15 days on April 30, and Attorney General Blanche committed in writing to meet the deadline before testifying at his July 15 confirmation hearing that the agencies were "working diligently with the intelligence community" ^[2][4].

The reason the deadline matters is the calendar on both sides of the fight. Section 702 itself lapsed this summer and collection continues under the court's March certification through next March ^[1]. The DNI has warned Congress that Section 702 would "go dark" unless lawmakers approved a clean reauthorization by November 20 ^[4]. Elizabeth Goitein of the Brennan Center disputes the deadline: "The real question is when the government has to submit its applications, and that is the effective deadline for when Congress has to act," she said, putting the filing date at March 17, 2027 ^[4]. The coalition's ask is blunt. Congress should "refuse to support any reauthorization of Section 702" while the opinion is withheld, and should "reauthorize Section 702 only with a warrant requirement" ^[1]. Wyden and Schiff's framing was equally direct: "As Congress continues to debate the fate of this authority, lawmakers must be able to discuss these abuses openly," and "Otherwise, the American people will rightly question whether a government that so flagrantly violates basic legal requirements" ^[1]. The Senate's lame-duck session begins November 9 ^[2]. The site's full write-up on the withheld ruling and the SAFE Act explainer cover the reauthorization mechanics in more detail.

GCHQ Lost Thousands of Drives and Cannot Say How Many

Connor Jones at The Register reported on October 30 that Britain's signals intelligence agency cannot account for "thousands" of removable storage devices, including USB sticks, external hard drives, compact discs, and memory cards, and that the loss has been known to officials for years ^[5][6]. IPCO described the missing media as "a very significant number," and Alexander Martin at The Record reported the same day that the watchdog was "very concerned" and "disappointed with the time it has taken GCHQ to investigate" ^[5][6]. Commissioner Sir Brian Leveson said GCHQ "has not yet established a sufficiently clear understanding of the scale, location, and recoverability of lost removable media" ^[5][6]. GCHQ said its removable-media management system had not been good enough to track the items registered for use, said it had acted on IPCO's recommendations, and said an automated asset-tracking system due by the end of 2026 will let it report a precise figure ^[5].

The point of the reporting is what the drives held. GCHQ is the agency that handles bulk intercepted communications, and the same watchdog report raises concerns about the safeguards for overseas partners accessing bulk intercepted data ^[5]. GCHQ declined to say whether unauthorized access to the missing media is known, and The Record notes that IPCO's report does not allege that adversaries obtained the devices ^[6]. Nobody has produced evidence the drives reached the wrong hands. Nobody has produced evidence that they did not either, and after several years the agency cannot tell you which question to ask.

The same report contains two other findings that deserve more attention than they got. The Register reported that a bug in a web application used for target discovery led to five port stops under the Terrorism Act, causing what IPCO called "serious prejudice" to the people affected, and that IPCO raised concerns about the safeguards for overseas partners accessing bulk intercepted data, with inadequate arrangements for one partner's access to journalistic material ^[5]. The report also covers Agent X, the MI5 informant case, and Leveson's assessment: "the case of Agent X is as serious as anything I have dealt with as Commissioner" ^[5]. A spy agency that cannot count its own drives is in an awkward position to police a partner's access to the same data.

GM Pays $220 Million Over Driver Tracking in 35 States and DC

Brandon Vigliarolo at The Register reported on October 30 that GM agreed to pay $220 million to 35 states and the District of Columbia to settle allegations that the company collected and shared connected-vehicle driving data without drivers' informed consent ^[7]. The program at issue is OnStar's Smart Driver, which collected identifiable data including precise location, speed, seatbelt use, and other driving metrics, then shared it with data brokers. Insurers bought that data and used it to set premiums ^[7]. Investigators said GM induced enrollment through dealerships, sometimes accepted terms on customers' behalf, and failed to notify many drivers. More than 370,000 people were told through the OnStar app that they were not enrolled ^[7].

The settlement is not just a check. GM must limit the driving data it collects, improve protections for customer personal information, and stand up a "comprehensive data privacy program" ^[7]. It builds on earlier orders: a Federal Trade Commission settlement finalized in January 2026 that requires express consent for covered collection, use, or sharing of connected-vehicle data, plus access, deletion, and geolocation-disabling options, and a separate California settlement for $12.75 million in May 2026. A Texas case is still running ^[7]. GM discontinued Smart Driver in April 2024 ^[7]. A spokesperson said the agreement "reinforces steps we've taken to strengthen our privacy practices" ^[7].

This is the same chain the California CCPA penalty story followed earlier this year, now scaled up and settled. The arithmetic is what makes it more than a privacy-policy failure. A single car is a data point. A fleet of them, collected through a customer account and sold to insurers as a way to price risk, is a surveillance system built out of ordinary hardware, and the people inside it never got to say yes or no. The on-paper remedy here is consent, which is exactly the thing the FISA fight is about and exactly the thing that keeps not arriving.

EFF Gets 800 Pages on the Medicare AI and the Safeguards Do Not Hold Up

Lena Cohen and Matthew Guariglia reported at EFF on October 27 that the organization obtained and released more than 800 pages of records on the Centers for Medicare and Medicaid Services AI model used to evaluate Medicare prior-authorization requests, called WISeR, the Wasteful and Inappropriate Service Reduction model ^[8]. EFF sued the government in March 2026 for the records. The release includes requests for information and corrective action plans involving at least four of the six selected vendors, along with vendor responses, applications from all 66 applicants, and the task orders and contracts for evaluation, implementation, and monitoring ^[8]. The documents also show a rushed rollout: CMS was sending RFIs and corrective action requests during the program's first three months, and reported problems included turnaround times beyond guidelines, potentially inaccurate denials, inadequate customer service, technical failures, and poor provider communication ^[8].

The vendor side of the story is worse than the schedule side. One vendor, Innovaccer, received two corrective-action-plan requests and a non-compliance notice over an incomplete provider portal, and said CMS had been warned it might not be ready by launch. Another attributed problems to the "compressed implementation timeframe" ^[8]. The claims about human oversight are the part EFF says cannot be checked. Vendor applications assert human review, monitoring, and audit trails, but the detail needed to evaluate those claims was redacted or missing. CMS promises that "Non-affirmations will require the review of a human clinician," EFF reports, but reviewers may lack the time or the authority to override the model ^[8]. The applications cite provider behavior, demographics, utilization-management datasets, performance metrics, and historical claims, without disclosing enough about how that data is used ^[8].

The financial incentive closes the loop. EFF argues that the way CMS pays the vendors gives them "an incentive to deny as many claims as possible" ^[8]. An algorithm that flags a claim for denial, a vendor paid per decision, and a human reviewer who cannot realistically read every file is a system built to deny. EFF's asks are transparency, real oversight, and legal limits on automated decision-making in Medicare ^[8]. The site's AI-agent breach coverage and the history of American surveillance state explainer cover the other half of the same problem: automated decisions that nobody in the chain can explain.

Ofcom's Shortlist for Age-Checking a Teen Social Media Ban

Connor Jones at The Register reported on October 29 that the UK's communications regulator Ofcom kept four "highly effective age assurance" methods for the proposed ban on social media for under-16s, and reserved three more for adults ^[9]. The four are photo-ID matching, banking checks, digital ID services, and facial age estimation. On the banking method, Ofcom estimates 90 percent of UK 16- to 17-year-olds have a bank account. On facial age estimation, Ofcom says the "margin of error of a few years" means some users may need a second check ^[9]. The adult-only fallbacks are credit-card checks, mobile-network checks, and email checks, and Ofcom says "many" adults will be verified even when age can be inferred from previous activity ^[9].

The identity requirement underneath all four methods is the part to sit with. Ofcom estimates 94 percent of UK 16- to 17-year-olds hold acceptable identification such as a passport, driving licence, or Proof of Age Standards Scheme card, and that roughly 150,000 have neither ^[9]. That means a meaningful slice of the country will need to go to a third party to prove they are old enough to use a service they are legally required to use. Big Brother Watch director Silkie Carlo called the proposal "a national ID system by the back door" and warned it locks out people without ID and people unwilling to hand identity documents to foreign technology companies ^[9].

The timing is set: the ban is expected to take effect "next year," the government was expected to publish the first regulations by the end of 2026, and Ofcom plans close monitoring for the first 12 months after enforcement ^[9]. Ofcom's own chief executive, Dame Melanie Dawes, compared the circumvention risk to the way motivated teenagers get around the alcohol rules, and Ofcom acknowledged that "the risk of circumvention cannot be entirely eliminated" ^[9]. Accepting that children will find a way around age checks does not make the age checks smaller. It makes them a permanent collection of identity documents by people who did nothing wrong, to enforce a rule aimed at people who did. The site's age-verification explainer and UK VPN age-gate coverage track the same infrastructure from the other end.

AI Agents Are Driving the Attacks Now, and the Paper Trail Is Better Than Ever

Microsoft's security team reported on October 21 that a group it tracks as Storm-0940, described as "a China-state aligned actor (with no known industry aliases)" that has been active since 2021, used an open-source penetration-testing framework called ARTEX to compromise ConnectWise's Cloud Backup service and then attempt sign-ins against more than 2,250 Microsoft 365 tenants ^[10]. Microsoft describes ARTEX as an open-source penetration testing framework that can use any AI model to automate a variety of offensive security workflows, running from reconnaissance through data exfiltration ^[10]. Matt Kapko at CyberScoop quoted Microsoft describing ARTEX as "a system of LLM-driven agents that run an autonomous penetration test from reconnaissance through data exfiltration," and noted Microsoft's warning that "multiple threat actors from various regions are using ARTEX to strengthen capabilities and speed up the tempo of attacks" ^[11].

The numbers in the Microsoft post describe a machine working at a pace no human team matches. A password-spray campaign ran October 10 through October 14 against 327 accounts across eight organizations, producing more than 151,000 failed logons, peaking on October 11 at 2,900 failures per minute with more than 111,000 failures concentrated into a single day. Downstream abuse of the compromised applications began October 15. In some attacks the actor moved from initial targeting to privileged cloud access in less than three hours ^[10]. What Microsoft will not say is how much was actually taken: a successful authentication "does not necessarily indicate that customer data was accessed or exfiltrated," and the activity "did not result from a compromise of or a vulnerability in Microsoft corporate systems, products, or services" ^[10].

ARTEX is not a one-off. On October 28, Connor Jones reported that the UK domain registrar 123 Reg disclosed an attack that began September 18, calling it "the first known AI-driven cyberattack against our company." The firm found "AI cyberattack 'skills' hosted on Chinese infrastructure, including copies of skills open sourced by Anthropic and several other AI exploitation frameworks and scripts," and assessed attacker infrastructure hosted in China and Singapore as "indicating a strong Chinese nexus." The company could not rule out that the intruder copied the data: "it's possible they created copies of the information" ^[12]. Meanwhile the human supply chain kept serving up the keys. Jessica Lyons reported on October 26 that LiteLLM traced its supply chain intrusion to a master key committed to a public GitHub repository on May 27, 2026 and still valid on October 10, when an attacker used it to extract a GitHub CLI token and reach owner-level access across the project's repositories. A poisoned version 1.107.0 uploaded to PyPI on October 21 pulled roughly 49,000 downloads before quarantine ^[13]. And on the same day, Jessica Lyons reported that JFrog researchers had found a worm arming 42 real GitHub repositories belonging to 11 owners, with execution hooks targeting Claude, Gemini, Cursor, and VS Code and 12.9 to 13.7 MB polymorphic wrappers, which JFrog said were "large enough to evade the GitHub search and API paths normally used to find this kind of compromise" ^[14].

The defensive side of the ledger moved too. Cisco open sourced Project Quarry, a behavior-detection engine for AI agents that compares whole sessions instead of isolated actions, so a read of a sensitive file followed by an outbound send reads as one risky sequence ^[22]. The point for readers is the shape of the attack surface. These are not zero-days. They are ordinary keys in ordinary repositories, ordinary trust relationships between a backup vendor and its customers, and ordinary build pipelines, all pulled together by agents that work far faster than the humans watching ^[10][13][14]. The site's ConnectWise and ARTEX write-up goes deeper on the tenant exposure.

Wyden and Markey Ask Meta What Happens to Intimate Images the Glasses Caught

Samantha Cole reported for 404 Media on October 29 that Sen. Ron Wyden and Sen. Edward Markey wrote to Meta on Wednesday, October 28, about the company's AI glasses and the capture of people who never agreed to be recorded ^[15]. The letter asks Meta to explain its safeguards against nonconsensual capture, disclosure, or processing of sensitive imagery, whether it evaluated voyeurism and intimate-image risks, what imagery is used for AI training, and whether captured biometric and other data can be deleted ^[15]. Wyden and Markey's framing: "Human review to fulfill a service knowingly requested by the person depicted is one thing," and treating bystanders who never asked for anything as strangers to be reviewed and trained on is another ^[15].

The underlying reporting is what makes the letter hard to wave away. Cole notes that an October 2025 investigation found Meta Ray-Ban glasses used to secretly record harassment of massage-parlor workers, with Instagram accounts holding hundreds of thousands of followers posting videos of men entering parlors and soliciting workers for sex work. The BBC reported in January on a woman who was "bombarded" with calls after her number appeared with a video ^[15]. More than 70 people sued Meta earlier in 2026 alleging intimate images were shown to AI-labeling workers, and in March, Kenyan workers were reported reviewing uncensored footage, including sexual and violent material, with the contract ending in under two months ^[15].

The wearable angle is the surveillance story underneath the legal one. A camera on a face that nobody can see is a camera that nobody can consent to on the record. The site's group letter on Meta's smart glasses and EDPB smart-glasses report coverage track the same product from the regulatory side.

Quad9 Would Rather Leave France Than Publish a Block List

Joe Mullin reported at EFF on October 27 that Quad9, the Swiss nonprofit DNS resolver, was scheduled to stop serving France on November 24 after a court order required it to block websites accused of infringing a sports broadcaster's copyrights ^[16]. EFF reports about 20 additional lawsuits in France seeking comparable orders ^[16]. The order left Quad9 with two bad options, and the company described them plainly: block the sites globally, which would "extend French law to the whole world," or try to block only for French users, which EFF says Quad9 considered unreliable and feared "massive fines" for if it failed to block every target perfectly ^[16].

Think about what a DNS resolver is. It translates a domain name into an address, which means it sees every destination you reach and every one you try to reach ^[16]. A block order turns that resolver into a censorship filter that the government can aim at anything. Quad9 also blocks malware and phishing domains, so the same infrastructure that filters out badness filters out whatever a court says ^[16]. EFF's conclusion is that withdrawal is the lesser evil compared with "making promises it can't keep, or abandoning users' rights worldwide," and that compulsory blocking can be repurposed by "repressive governments" to censor speech globally ^[16].

EFF also used the story to argue that Congress should reject the pending site-blocking bills in the US ^[16]. The mechanism matters more than the verdict. Blocking is cheap to order and expensive to make correct, because a block list is a list of names. A resolver that cannot guarantee completeness cannot promise compliance, and a company that cannot promise compliance becomes a liability the moment a court's list grows past what its filters can track.

Threads Still Running From the Week

A federal court ruled that OPM's data grab for DOGE was unlawful. Mario Trujillo reported at EFF on October 20 that Judge Denise Cote of the U.S. District Court for the Southern District of New York issued summary judgment on October 20 holding that the Office of Personnel Management broke two Privacy Act provisions by disclosing records of millions of federal employees to DOGE agents, once between agencies and once by failing to protect the security and confidentiality of the records. Cote wrote that "OPM's decision to grant DOGE Agents broad access to its systems of recordings containing [personally identifiable information (PII)] has been held unlawful" ^[19]. The remedy is still open. The site's DOGE data-grab coverage tracks the rest of the thread.

EFF is asking the DC Circuit to curb grand jury subpoenas aimed at social media users. F. Mario Trujillo reported on October 28 that in a sealed appeal before the U.S. Court of Appeals for the District of Columbia, DHS served Reddit with an administrative subpoena for an anonymous user's identity, withdrew it after a challenge, then obtained a grand jury subpoena for substantially the same information ^[17]. The brief says such subpoenas, which are "issued by a prosecutor to gather evidence to present to a grand jury before any criminal charges is filed," "must receive proper First Amendment scrutiny" ^[17]. It was filed in early September 2026 and later unsealed, and it is signed by the ACLU of DC, EFF, the Center for Individual Rights, FIRE, the Cato Institute, and NetChoice ^[17].

EFF published the infection chain behind last week's spyware story. Bill Budington and Thorin Klosowski wrote on October 26 that Citizen Lab's AdFection research traced what happens after a Discover-Tech ad loads on a targeted device: the ad's JavaScript contacts Discover-Tech's "NOAS server," which receives the cookie-set advertising ID and IP address, recognizes the device as belonging to Al Jazeera, and returns a Candiru payload that infects the machine automatically. EFF calls it "a horrifying new malware campaign" and says the novelty is in the combination, "combining them, creating a highly targeted, extremely dangerous attack" ^[18]. The recommendations are consumer-level: block ads network-wide, reset the advertising ID, use Privacy Badger or a privacy-protective browser ^[18]. The site's October 18 roundup has the larger Discover-Tech and Candiru story.

Proton's CEO cancelled a company-wide AI interrogation. Joseph Cox reported for 404 Media on October 30 that Proton CEO Andy Yen called off nearly company-wide "re-interviews" on Friday, October 30, after 404 Media's October 29 investigation and a wave of internal discussion. The interviews would have had employees account for their AI use, with the apparent possibility of dismissal if leadership judged their usage insufficient ^[20]. Cox reports that in Proton's security Slack channel, an employee urged the company to spend several days penetration-testing new code before release, and Yen replied at the end of August that it was "bullshit and loser thinking" and told the channel to "stop lovingly handcrafting all our code and use AI more." CTO Bart Butler said employees had "We fucked up, we know we fucked up" ^[20]. Yen said the interviews had achieved their objective because employees were now talking about and learning about AI, acknowledged a personal communication failure, and said "Going forward, I'm still going to push" ^[20]. AI-use measures are to move into performance reviews, and the Proton Foundation board includes Yen, Sir Tim Berners-Lee, Carissa Veliz, Dingchao Lu, and Antonio Gambardella ^[20].

Two stories from earlier in October are worth re-reading next to this week's reporting. The site's eMudhra certificate-key piece covered eight compromised certificate intermediates, and the Livonia police AI-generated child photos story covered synthetic imagery used in an undercover sting.

The Week Ahead

November 6. That is the deadline Wyden and Schiff set for the administration to explain the withheld FISA opinion and name the officials responsible for the delay ^[21]. A letter that names nobody settles nothing, and the attorney general's required statement of the opinion's significant legal conclusions has still not appeared ^[4].

November 9. The Senate's lame-duck session begins ^[2]. Any attempt to attach a Section 702 renewal to end-of-year legislation has to survive a Senate that has just heard the FBI director testify to 99 percent compliance ^[3].

November 20. The "go dark" deadline the DNI pressed on Congress ^[4]. Goitein's counter is that the real filing deadline is March 17, 2027 ^[4], so the interesting signal is which date the administration actually uses in public.

November 24. Quad9 stops serving France ^[16]. Whether the other DNS resolvers and French providers follow, and whether the roughly 20 related lawsuits produce orders that are harder to comply with, is the test of whether this is an isolated decision or the start of a European pattern.

End of 2026. Two deadlines land together: Ofcom's first regulations for the under-16 social media ban ^[9], and GCHQ's automated asset-tracking system, which the agency says will let it finally say how many drives went missing ^[5]. The GCHQ number is the one worth watching. If it comes back in the low hundreds rather than the thousands, the headline was the loss of the paperwork, not the loss of the data.

Still open on the CMS front. EFF's remedy demands and the unresolved questions about human review ^[8] need a response from CMS, and some of the vendor records EFF obtained remain redacted.

And the intrusion cleanup. Microsoft's guidance for ConnectWise customers was to keep affected connections disabled until ConnectWise issues formal restoration instructions, and Microsoft cautions that a successful authentication does not mean customer data was accessed, so what the agents actually touched is not settled ^[10][11].

Sources

1. CyberScoop, Tim Starks: "Senators, rights groups say Trump administration breaking law by withholding surveillance court opinion" (October 26, 2026) (the March 17 FISC opinion, the letter from Wyden and Schiff to Director of National Intelligence Jay Clayton and Attorney General Todd Blanche, the coalition demand that Congress refuse reauthorization and add a warrant requirement, the summer lapse of Section 702, and collection continuing under a court ruling through March)
2. Center for Democracy and Technology: "CDT Demands Release of Secret Court Opinion" (October 26, 2026) (the declassification procedure under 50 U.S.C. Section 1872, the 180-day period, the September 13 deadline, the required attorney general statement of significant legal conclusions, the signatory organizations, the April 30 Cotton and Warner request, and the November 9 start of the lame-duck session)
3. Just Security, Jake Laperruque: "Why is the Government Hiding Court Rebuke on Warrantless FISA Surveillance?" (October 29, 2026) (the effect of the filter tools on audits and on the annual query count, the accounts of Wyden and Massie who read the opinion, and FBI Director Kash Patel's testimony on query compliance)
4. The American Prospect, James Baratta: "Trump Administration Ignores Statutory Deadline to Declassify FISA Court Opinion" (September 24, 2026) (the DNI warning that Section 702 would "go dark" by November 20, Elizabeth Goitein's account of the March 17, 2027 filing deadline, Donald Bell of the Project On Government Oversight, and Attorney General Blanche's July 15 confirmation hearing testimony)
5. The Register, Connor Jones: "British intelligence agency misplaced 'thousands' of storage devices, years later still can't say how many" (October 30, 2026) (the missing removable media, Commissioner Sir Brian Leveson's assessment, GCHQ's response and its asset-tracking system due by the end of 2026, the five port stops under the Terrorism Act, the overseas partners accessing bulk intercepted data, and the Agent X case)
6. The Record, Alexander Martin: "GCHQ cannot account for thousands of data storage devices" (October 30, 2026) (the types of missing media, IPCO's criticism of the time GCHQ took to investigate, GCHQ's refusal to say whether it knew of unauthorised access to the media, and the absence of any allegation that adversaries obtained the devices)
7. The Register, Brandon Vigliarolo: "GM cracks the petty cash box for multistate driver tracking settlement" (October 30, 2026) (the $220 million payment to 35 states and the District of Columbia, the OnStar Smart Driver program, the data brokers and insurers, the 370,000 customers told they were not enrolled, the January 2026 FTC settlement, the $12.75 million California settlement, and the April 2024 discontinuation of Smart Driver)
8. EFF, Lena Cohen and Matthew Guariglia: "New Records Obtained by EFF Expose More Problems with Medicare's AI Experiment" (October 27, 2026) (the WISeR model, the more than 800 pages of records released after EFF sued in March 2026, the six selected vendors from 66 applicants, the corrective action plans including Innovaccer's two requests, the unverifiable human-review claims, and the vendor incentive to deny claims)
9. The Register, Connor Jones: "Shortlist of age checks for the UK's under-16 social media ban is out, 'some' adults will need verifying" (October 29, 2026) (the four retained age-assurance methods, the three reserved for adults, the 94 percent identification rate among UK 16- to 17-year-olds, the 150,000 without acceptable ID, the expected start of the ban, and Big Brother Watch director Silkie Carlo's objection)
10. Microsoft Security Blog: "Storm-0940 uses ARTEX agentic AI in ConnectWise supply chain compromise" (October 21, 2026) (the description of Storm-0940, the ARTEX framework, the more than 2,250 tenants, the 327 accounts across eight organizations, the 151,000 failed logons peaking on October 11, the October 15 downstream application abuse, and Microsoft's customer remediation guidance)
11. CyberScoop, Matt Kapko: "Microsoft attributes ConnectWise supply chain attack to China-linked Storm-0940" (October 21, 2026) (the ARTEX description, Microsoft's warning about other threat actors using the framework, and the sign-in rate against the more than 2,250 tenants)
12. The Register, Connor Jones: "123 Reg discloses 'AI cyberattack' on customer data, suspects China nexus" (October 28, 2026) (the September 18 attack on the UK domain registrar, the "first known AI-driven cyberattack" claim, the China nexus assessment, the AI exploitation skills on Chinese infrastructure, and the CrowdStrike report CSA-261162)
13. The Register, Jessica Lyons: "LiteLLM supply chain attack started with an exposed master key" (October 26, 2026) (the master key committed to a public GitHub repository on May 27, 2026, the October 10 use of the key, the poisoned version 1.107.0 on October 21, and the roughly 49,000 downloads before quarantine)
14. The Register, Jessica Lyons: "Worm armed 42 GitHub repos by hiding a payload too big to fail" (October 26, 2026) (the JFrog research, the 42 repositories and 11 owners, the Miasma "testmetest" npm worm, the execution hooks for Claude, Gemini, Cursor, and VS Code, and the polymorphic wrappers sized to evade GitHub's own search and API paths)
15. 404 Media, Samantha Cole: "Senators Demand Meta Answer for 'Pervert Glasses' Following 404 Media Investigation" (October 29, 2026) (the October 28 letter from Wyden and Markey to Meta, the October 2025 reporting on the recording of massage-parlor workers, the more than 70 lawsuits filed this year, and the March reporting on Kenyan contractors reviewing footage)
16. EFF, Joe Mullin: "French Court's Site-Blocking Orders Drive Quad9 Out Of France" (October 27, 2026) (the order to block sites, the roughly 20 additional lawsuits in France, the November 24 withdrawal date, Quad9's two unacceptable options, and EFF's warning about the reuse of blocking orders by repressive governments)
17. EFF, F. Mario Trujillo: "Court Should Curb Abuse of Grand Jury Subpoenas Targeting Social Media Users" (October 28, 2026) (the sealed appeal before the DC Circuit, the withdrawn DHS administrative subpoena and the later grand jury subpoena, the September 2026 amicus filing, and the six signatory organizations)
18. EFF, Bill Budington and Thorin Klosowski: "Lessons Learned From 'AdFection' and What To Do Next" (October 26, 2026) (the Citizen Lab trace of the Discover-Tech infection chain, the NOAS server, the advertising ID and IP address, the Candiru payload, and the consumer-level protections EFF recommends)
19. EFF, Mario Trujillo: "Victory! Court Rules OPM's Disclosure of Personal Records to DOGE Unlawful" (October 20, 2026) (the October 20 summary judgment by Judge Denise Cote in the Southern District of New York, the two Privacy Act violations, and the unresolved remedy)
20. 404 Media, Joseph Cox: "Proton's CEO Cancels AI 'Re-Interviews'" (October 30, 2026) (the cancelled company-wide interviews, the Slack messages quoted in the story, CTO Bart Butler's response, the Proton Foundation board, and the plan to move AI-use measures into performance reviews)
21. State of Surveillance: "Government Withholds Secret FISA Ruling on Warrantless Spying" (October 30, 2026) (the November 6 deadline set by Wyden and Schiff for the administration to explain the delay and name the officials responsible)
22. The Register, Jessica Lyons: "Cisco open sources tool for detecting AI agents on hacking sprees" (October 28, 2026) (Project Quarry, the session-level behavior detection, and the July 10, 2026 Hugging Face incident in which an agent validated exposed tokens)