Week of November 16 to 22, 2026:

  • ICE is paying AT&T millions for raw phone data and the analysts to sift it. The procurement document says the deal gives the agency "access to AT&T's raw data and experienced analysts", and the paperwork runs on administrative subpoenas that no judge approves [1].
  • The FBI's National Crime Information Center was breached. The database holds the warrant and background-check records officers read at traffic stops, and nobody has said what was taken or whether the data was altered [2][3].
  • EFF and more than 25 civil-liberties groups asked Congress to keep Section 702 out of the defense bill. The authority expired in June 2026 and the program is set to turn off within months [4].
  • A UK spy tribunal made the government admit the secret technical capability notice served on Apple. The three judges said "That does not mean every detail must be kept secret in order to preserve national security" [5].
  • Nearly two million test prompts showed AI providers and resellers ignoring the privacy flag users set. OpenRouter removed a zero-data-retention label from MiniMax endpoints [6].
  • Therapy notes landed on the London Care Record, which up to 140,000 NHS staff can open. The trust admitted that "key session content" was uploaded automatically, and a junior health minister said the sharing rests on implied consent [7].
  • The Dutch watchdog tested six AI camera glasses and told buyers to stay away. On virtually all of them the recording light was dim and easy to cover [8].
  • Signal now blocks operating-system AI agents from screenshotting messages. Signal cannot do the same on macOS, and the Dutch regulator found recording lights on smart glasses that were dim and easy to cover [9][8].

Also this week: The FCC authorized state and local prisons to jam contraband cell phones for the first time, and the industry's own pilot test found 911 calls blocked and interference reaching two and a half miles [11]. A study of 81,000 US census tracts and 5,291 data centers found that majority-Black neighborhoods have 2.2 times the odds of hosting a data center than a white neighborhood in the same state. The researchers recommend data-center-specific zoning rules and racial impact statements before construction [13]. And federal prosecutors indicted Igor D'iu, 25, a Russian national accused of photographing employees of a Silicon Valley technology and defense company and following one of them for nearly 30 miles [14].

Nobody Signed Off

Read this week's stories one at a time and they look like a pile of unrelated bad news. Read them as one beat and the shape is obvious. In every case a safeguard existed. It had been written down, tested, and posted on a page somewhere. Then it turned out that nobody with the authority to enforce it ever had to. ICE is buying raw phone data and the analysts to sort it, and the legal paperwork that authorizes it is an administrative subpoena an agency issues to itself without a judge. Congress is being handed a Section 702 renewal attached to the defense bill so the debate gets skipped rather than held. You can set a zero-data-retention flag on an AI prompt and the provider can ignore it. Your NHS trust can upload your therapy notes to a record 140,000 staff can open, and your consent is implied rather than asked for. The light on the camera glasses that records you is dim enough to cover with a thumb.

The counterpoint to all of that sits at the bottom of the stack, in a ballot box. Voters in York, Maine, backed a measure to remove the town's automated license plate readers and to require voter approval for any future installation, with nearly three-quarters of York voters supporting the ban [12]. Voters in League City, Texas, rejected continued police use of the cameras, and the result, formally approved on November 17, opened a 72-hour removal window. In Pflugerville, Texas, Proposition B passed and will require the city charter to spell out how surveillance technology gets used [12]. That is what a check that binds looks like: slow, local, and on the record. Everything else on this list works the other way around.

There is a third thread, and it is about who carries the weight. A study by researchers at George Washington University and the Center for Security, Innovation & New Technology (CSINT) at American University analyzed 81,000 US census tracts and 5,291 data centers, and found that majority-Black neighborhoods have 2.2 times the odds of hosting a data center than a white neighborhood in the same state, 2.6 times in majority-Hispanic neighborhoods and 4.2 times in majority-Asian neighborhoods. The pattern held after controlling for population, household income, education, land costs, proximity to major transmission lines and state fixed effects. Co-author Dhanaraj Thakur said: "It's just how embedded race is with all these other factors". The researchers recommend data-center-specific zoning rules and racial impact statements before construction [13]. The infrastructure for the AI economy is being sited somewhere. The people who live there did not get to vote on it.

ICE Is Buying AT&T's Analysts to Dig Through Raw Call Records

Joseph Cox reported for 404 Media on November 19 that ICE is paying AT&T millions of dollars for access to large quantities of raw AT&T data and for AT&T's own analysts to sift through it. The reporting rests on an internal ICE procurement document obtained through a public records request, along with other procurement records that experts reviewed. The document reads: "This requirement provides ICE access to AT&T's raw data and experienced analysts" [1].

The product is called Data Analytics and Support Services, also called Data Analytical Services, and it appears to be related to Hemisphere, the earlier program in which AT&T provided US authorities at state, local and federal levels access to AT&T user data before it shut down. Civil-liberties experts who reviewed the records call it a mass surveillance program, and they say the data can include trillions of phone records and location data. Because AT&T carries calls for other carriers, people who are not AT&T customers are likely swept in too. AT&T told 404 Media that the data is made available under a valid legal order [1].

The weak point is who signs that order. The experts point out the arrangement runs on administrative subpoenas, which agencies issue themselves and which do not require a judge's approval. That is the difference between a targeted investigation and a mass search. Because no judge reviews the paperwork, the scope of what ICE can pull is a matter of internal practice, and the experts who read the records say this raises concerns that ICE could use the information for essentially any purpose it chooses [1].

The FBI's Crime Database Was Breached and the FBI Is Investigating

Evan Perez reported for CNN on November 20 that apparent hackers had intruded into the National Crime Information Center, the FBI's nationwide criminal information database, which law enforcement agencies across the country use to share information on warrants and background checks. It is the second such breach of FBI systems in recent months. The FBI has not publicly confirmed the incident and did not respond to CNN's request for comment, but two people briefed on the issue said the agency is investigating [2].

Lawrence Abrams reported for BleepingComputer on November 21 that the National Crime Information Center is an FBI-operated database used across the country to share information about wanted and missing persons, outstanding arrest warrants, stolen vehicles and firearms, and protection orders, and that police officers routinely use the system during traffic stops, criminal investigations and background checks. It remains unclear whether the attackers accessed sensitive law enforcement records or whether the intrusion affected the system's availability, and there is currently no indication the incident is connected to ShinyHunters or the earlier breach [3]. A September attack by the ShinyHunters group targeted a third-party platform associated with the FBI's recruitment website, FBIJobs.gov [2]. The resulting FBI investigation led to arrests of suspected ShinyHunters members, including a Canadian cybersecurity executive reportedly linked to the attacks [3].

The fear is not only theft. A former FBI official who led investigations into cybercriminal groups told CNN: "Every day, officers across the country rely on NCIC to identify wanted fugitives, check for outstanding warrants, and access information that can directly impact officer and public safety. My concern isn't simply whether hackers gained access to sensitive information, although that is an issue, but whether they could compromise the accuracy of the information that officers rely on." Stolen records can be exposed. Altered records are worse, because the officer at the window cannot tell the difference between what the file says and what somebody wrote into it [2].

The president's response is its own story. On Saturday, Trump told Fox News in a telephone interview conducted by his former White House press secretary Kayleigh McEnany: "But maybe we'll learn things that we would have never learned otherwise because, you know, you had the Russia, Russia, Russia hoax." He also said: "Maybe we'll learn about the voter fraud, all the voter fraud that took place," without citing evidence. Trump called his own framing "I always like to put things in a positive spin" [2]. On Thursday, after bureau officials learned of the incident, FBI Director Kash Patel used a Fox News appearance to tout the bureau's cybersecurity work, referring to Operation Winter Shield as if it were still in the works and saying: "Other administrations could have made these updates, but they didn't," [2]. The bureau is investigating the breach of its own database on the same week its director was on television promoting its defenses.

Section 702 Is Being Shoved Into the Defense Bill

Section 702 of FISA expired in June 2026 with no agreement among lawmakers on the terms of its reauthorization, and the program is set to truly turn off within the next few months. Matthew Guariglia reported at EFF on November 20 that EFF and more than 25 civil-liberties organizations, including the ACLU, Color of Change and the Brennan Center, wrote to Congress asking lawmakers not to let a renewal ride along on a must-pass vehicle such as the National Defense Authorization Act or a rushed end-of-year funding bill [4].

The letter's argument is procedural, and that is the whole point. It reads: "The NDAA is among the most important bills Congress considers each year. It should not be used as a vehicle to circumvent or to cut short this long-running bipartisan debate over Americans' civil liberties and constitutional rights, nor to force lawmakers to choose between supporting our national defense and ending warrantless spying on Americans. Adding language in the NDAA conference to reauthorize Section 702 that has not been fully debated and voted on would be a serious procedural and substantive failure, and a betrayal of the democratic process" [4].

Why the procedural fight matters is the mechanism underneath it. Section 702 lets the government compel a telecommunications company to hand over data to the National Security Agency for monitoring communications overseas. Because international and domestic traffic interconnects, communications between people abroad and people in the United States get swept up as well, and once that data lands here the FBI and several other agencies search U.S. persons' data without a warrant. EFF calls that unconstitutional and says it creates a potent tool for anyone who wants to punish people by identity or belief [4]. The difference between a debated renewal and a buried one is the difference between a law that expires and a law that renews itself. Our FISA 702 reform scorecard tracks what each proposal would change, and our complete guide to the fight covers the mechanics in more detail.

A UK Spy Tribunal Made the Government Say It Out Loud

The Investigatory Powers Tribunal has confirmed that the UK government served Apple with a secret Technical Capability Notice, rejecting the Home Office's argument that acknowledging its existence would damage national security. Connor Jones reported for The Register that the three judges found that extensive media coverage, statements by US officials and government-authorized briefings had already made the notice public, and that the precise content of the notice and its supporting evidence remain undisclosed [5].

The judges did not let the government keep the whole file. The judgment says: "That does not mean every detail must be kept secret in order to preserve national security. There may be some information that could be disclosed publicly without damaging national security. It is not, however, practical to determine the precise boundaries now" [5].

A technical capability notice is issued under the Investigatory Powers Act 2016 and compels a company to make technical changes that facilitate the government's lawful surveillance powers. The exact demands have never been publicly confirmed, though the order has been widely reported as seeking access to end-to-end encrypted iCloud data. The tribunal confirmed the notice's existence, not its requirements. Apple's first challenge was dismissed after a change in circumstances, it filed a second claim in July 2026, and the substantive hearing is planned for the first half of 2027. Liberty and Privacy International are running separate proceedings against the notice regime [5].

The government spokesperson said the notice "simply requires that they maintain the ability to do so" and that "Decisions affecting the safety of the British public must be made by democratically elected governments and independent judges, not in the boardrooms of technology companies" [5]. Read that next to the ruling and the gap is hard to miss. The Home Office served the notice and kept even its existence secret until a tribunal ordered the admission. Our earlier coverage of the Apple backdoor fight tracks what the Home Office was demanding in the first place [5].

Your Zero-Data-Retention Flag Did Nothing

HD Moore, chief research officer at Dragos, assembled a collection of AWS access keys tied to empty accounts, service passwords, unique URLs, DNS names and private git repository references, then seeded each secret into exactly one prompt sent to a single provider and watched where they surfaced. Dan Goodin reported for Ars Technica on November 20 that Moore wrote: "We sent nearly two million AI prompts containing unique, trackable secrets to AI providers, then watched where those secrets surfaced" [6].

The findings included evidence that at least three AI providers were compromised by attackers who used their access to siphon the prompts they received, including requests the provider had rejected before any model ran. Resellers in the inference marketplace turned on request logging and waited, and some of the stolen requests were marked as zero data retention. Moore's summary of the lesson: "A privacy flag doesn't help when the attacker is already inside." The bait keys were used "tens of thousands" of times in AWS API calls [6].

The scraper is the part that should worry you most, because it needs no attacker at all. Moore wrote: "The clearest example involves a Chinese AI lab." He found that this lab's systems fetched the secret URLs from more than half of the prompts sent to its models between July and October, including roughly two-thirds of the requests marked as no data collection or zero data retention, and that the same scraper fired for prompts sent to US hosting providers reselling that lab's models. A customer who picked a US provider specifically to keep data in the US did not get what they paid for. One US hosting provider crawled a secret URL more than two months after a single inference call to its search endpoint. Moore's advice is to host models on your own servers if you can afford to, otherwise buy from a first-tier provider with "one contract and a privacy policy to read", and refuse default provider or reseller pools [6].

The November 21 update is what moved the story. OpenRouter sent customers an email saying that some of their prompts may have been read, that the potential exposure involved prompts sent to MiniMax, a China-based model provider, and that the researchers' planted URLs were accessed whether the prompts went to MiniMax directly or through three other resellers: AtlasCloud, Novita and StreamLake. OpenRouter has removed a zero data retention label from MiniMax-model endpoints [6]. A privacy setting that was wrong, taken away after the fact.

Your Therapy Notes Went to a Record 140,000 NHS Staff Can Open

Matthew Weaver reported for The Guardian on November 20 that a patient known as Rosanna found her psychotherapist's notes about grief and trauma on a computer screen during a visit to her GP. Freedom of information inquiries showed that staff who did not work at the trust where she was treated had accessed them. The notes had been uploaded to the London Care Record, an electronic patient record accessible to up to 140,000 NHS staff [7].

She stopped treatment for bipolar disorder at South London and Maudsley NHS trust, and was without mental health support during a close family bereavement. Rosanna described the discovery this way: "It's like discovering that your personal diary has been photocopied and placed in a library." She is concerned that sharing therapy notes, including those of children, could present a danger to patients, and said: "Where trauma is involved, it can create a safety risk" [7].

The trust said "no intimate details" from therapy sessions had been shared, but in a response it admitted that "key session content" from therapeutic discussions was automatically uploaded, and told her: "Very private material can be shared in the context of treatment". A subject access request she sent in September has gone unanswered. The Information Commissioner's Office confirmed it will investigate after the former Labour minister Philip Hunt asked it to expedite the case [7].

Junior health minister James Frith replied in parliament that all records "relating to direct care are shared on a basis of implied consent". Rosanna's answer: "It's implicit that therapy is confidential unless you're told otherwise, within the normal safeguarding limits. Patients should be told where their data is going". The fight over the single patient record proposed in the health bill is running straight into this one. Rosanna supports it but wants more: "Transparency over what will be on the shared record and the option to opt out is essential" [7]. Our coverage of the NHS Palantir platform covers the other fight over who gets to see NHS records.

You Cannot Tell That You Are Being Recorded

The Dutch Data Protection Authority tested six models of AI-powered smart glasses, priced from under 100 euro to more than 500 euro, and advised consumers against buying them. Connor Jones reported for The Register on November 20 that the authority's vice chair, Monique Verdier, said: "Glasses that film you unnoticed are already concerning." The recording light was dim and easy to cover on virtually all of the devices tested, and the glasses could be induced to record with the light covered. People nearby could sometimes connect to the device without permission, view recordings and operate the camera. The AI functions process data primarily on servers outside the European Union, and users were not given enough information about how the data was handled [8].

The safety failures were stranger than the privacy ones. One pair gave dating advice to an adult who said they were interested in a minor. During a cardiopulmonary resuscitation simulation, another told the wearer that emergency services had been called and were scheduled to arrive, when nobody had called them. Some mishandled safety warnings on children's toys, some misread medication leaflets, and some misinterpreted extreme-weather warnings. Verdier's conclusion: "With the camera glasses examined, that trust cannot currently be justified." Her advice to lawmakers was "It is important that we, as a society, learn to deal with these types of high-risk AI products" [8].

Signal ran into the same wall from the other side. Thomas Claburn reported for The Register on November 20 that Signal's iOS and Android apps now block Siri, Gemini and other operating-system AI agents from taking screenshots of Signal messages by default, because those agents can capture the screen and pull the text out with optical character recognition. Signal CEO Meredith Whittaker said: "In doing so, your Signal data is captured, then made available to these agentic systems in order to allow them to answer your questions and do your bidding." Her sharper complaint is aimed at the platform vendors: "Even where we are able to detect and block AI data access, Google, Apple, Microsoft, and others do not provide ways for developers to easily restrict such AI data access without making drastic compromises to the core user experience." Signal cannot implement the same protection on macOS, because Apple does not give developers the same tools [9]. Apple is adding finer controls for device control and data access in the macOS 27.3 public beta.

And Ring has announced an indoor drone. Matthew Guariglia and Bill Budington reported at EFF on November 16 that Ring's new indoor camera can follow preset routes through a home and be launched by another Ring sensor or alarm. EFF's objection is that the AI Pilot maps routes and models the interior of the home while the footage goes to company servers, that a warrant might give police not just footage but control of the drone itself, potentially letting them search a home without the resident's involvement, and that putting autonomous cameras in ordinary homes lowers what people expect from their own walls. EFF called the safety pitch "it sounds like security theater and marketing" [10]. Our coverage of Ring's facial recognition class action and of the smart-glasses push on Congress covers the same product line from the regulatory side.

The FCC Let Prisons Jam Phones, and Voters Said No to Flock

On November 19, the FCC voted to let state and local prisons use jamming technology against contraband cell phones, the first time it has allowed it. Jon Brodkin reported for Ars Technica that FCC Chairman Brendan Carr said: "Today, we are finally authorizing state and local prisons and jails to use targeted jamming against contraband cell phones for the first time". Carr said "The technology has advanced to the point where it can be focused with geographic precision. No facility will be required to do it" [11].

The industry's own pilot test says the blast radius is wider than the cell. CTIA, the wireless industry lobby, filed with the FCC on October 1 after running the test with the Virginia Tech Applied Research Corporation and the South Carolina Department of Corrections. The filing says the micro-jamming system "blocked 911 calls, interfered with public safety and other communications, and caused interference that impacted wireless network infrastructure up to two and a half miles away", and that the pilot showed "substantial gaps" in jamming the targeted contraband devices, with call-completion failure rates in outdoor inmate recreation areas of "one-half to two-thirds". CTIA's conclusion was that "current jamming technology seems to be both over and underinclusive at once" [11].

The safeguard is that facilities apply to the FCC and "show through testing that it stays inside the designated area and doesn't interfere with lawful communications outside the prison walls", and that nearby communities receive notice before operations start. FCC Democrat Anna Gomez, the only Democrat on the three-member commission, voted with the majority after safeguards were added, and she noted that "For the last 90 years, the Federal Communications Commission has prohibited jamming by non-federal actors and even prohibited jamming equipment to be introduced into the American market" [11]. Freedom of the Press Foundation chief of advocacy Seth Stern wrote to FCC Commissioner Anna Gomez's office that the FCC should have weighed the value of contraband phones to people documenting abuse inside prisons, pointing to footage from The Alabama Solution, which bolstered an ongoing Department of Justice investigation into Alabama prisons and was crucial evidence in a civil lawsuit [11]. The order is the check, and the check is applied by the body that wrote the rule.

Then the voters. Beryl Lipton reported at EFF on November 18 that York, Maine, voted to remove the town's existing ALPR cameras and to require voter approval before any future installation, with nearly three-quarters of York voters supporting the ban. In League City, Texas, voters rejected continued police use of ALPR cameras; that result was formally approved on November 17 and began a 72-hour removal window, after which the cameras were "bagged" or covered to stop them collecting data. In Pflugerville, Texas, Proposition B passed, requiring the city charter to be amended with "protections governing the collection, use, retention, and oversight of data, facial recognition, and surveillance technologies" and "a responsible artificial intelligence framework." EFF's conclusion: "Surveillance is a choice, not a mandate, and our communities have the power to refuse it" [12]. Our Flock cancellation tracker follows the wider wave of cities walking away from the cameras.

The Week Ahead

November 24. Igor D'iu, 25, a Russian national indicted on conspiracy to act as an agent of a foreign government, acting as an agent of a foreign government without notice to the attorney general, and interstate stalking, is due back in federal court. Dan Goodin reported for Ars Technica on November 19 that prosecutors say D'iu photographed employees of a large technology and defense company and their computer screens at a coffee shop near its Palo Alto office, kept a list of home addresses, positioned his laptop so it could be remotely accessed, and followed an employee for nearly 30 miles until she reached police. He used a falsified California ID, an altered US passport and a Social Security card in the name "Hugo Landry Johns". Reuters and Bloomberg identified the target company as Palantir; Ars Technica could not confirm that, and Palantir did not respond [14].

The defense bill. Section 702's future is now a question about process. Watch whether the NDAA conference picks up renewal language, and whether members who want a warrant requirement for FBI access to Americans' data can vote on it without having to vote no on all defense policy [4].

The next UK ruling. The tribunal has confirmed the existence of the technical capability notice but has not yet ruled on whether further details of it may be published. The substantive hearing is set for the first half of 2027, and the separate Liberty and Privacy International proceedings are still running [5].

The FCC's jammer comment period. The commission said it is seeking comment on next steps, including ways to close remaining loopholes on other frequencies and the possibility of quiet zones around correctional facilities [11]. CTIA's own numbers are the argument for writing those rules tightly.

The ICO's investigation. The Information Commissioner's Office has only confirmed that it will investigate the London Care Record case. Rosanna's subject access request from September is still unanswered, and the trust's review of how it documents psychotherapy is still running [7].

League City's clock. The 72-hour removal window opened on November 17, and the cameras have been bagged or covered to stop them collecting data. The question worth watching is whether they stay that way, and whether other Texas towns follow a vote that was formally approved rather than merely counted [12].

Sources

  1. 404 Media, Joseph Cox - ICE Is Paying AT&T to Dig Through a Mountain of Call Records (November 19, 2026). https://www.404media.co/ice-is-paying-at-t-to-dig-through-a-mountain-of-call-records-hemisphere-data-analytical-services/
  2. CNN, Evan Perez - Trump downplays breach into FBI's crime database while reviving complaints about agency's Russia probe (November 20, 2026). https://www.cnn.com/2026/11/20/politics/fbi-hacked-again-target-crime-database
  3. BleepingComputer, Lawrence Abrams - FBI reportedly hacked again, this time targeting nationwide police database (November 21, 2026). https://www.bleepingcomputer.com/news/security/fbi-reportedly-hacked-again-this-time-targeting-nationwide-police-database/
  4. EFF Deeplinks, Matthew Guariglia - EFF to Congress: Mass Surveillance Is Too Important to Shove into Must-Pass Bill (November 20, 2026). https://www.eff.org/deeplinks/2026/11/eff-congress-mass-surveillance-too-important-shove-must-pass-bill
  5. The Register, Connor Jones - Spy court blasts UK secrecy, forces government to admit Apple 'backdoor' demand (November 20, 2026). https://www.theregister.com/security/2026/11/20/spy-court-blasts-uk-secrecy-forces-government-to-admit-apple-backdoor-demand/5311292
  6. Ars Technica, Dan Goodin - Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced. (November 20, 2026). https://arstechnica.com/security/2026/11/a-leaky-ai-supply-chain-is-exposing-secret-user-data-at-an-alarming-rate/
  7. The Guardian, Matthew Weaver - Patient 'horrified' to find her sensitive discussions were shared on system used by more than 100,000 NHS staff (November 20, 2026). https://www.theguardian.com/society/2026/nov/20/patient-confidentiality-sensitive-discussion-records-shared-nhs-staff
  8. The Register, Connor Jones - Dutch data watchdog highlights safety issues with AI glasses (November 20, 2026). https://www.theregister.com/security/2026/11/20/dutch-data-watchdog-highlights-safety-issues-with-ai-glasses/5311345
  9. The Register, Thomas Claburn - Signal slams door on prying AI agents to protect privacy (November 20, 2026). https://www.theregister.com/security/2026/11/20/signal-slams-door-on-prying-ai-agents-to-protect-privacy/5311466
  10. EFF Deeplinks, Matthew Guariglia and Bill Budington - Amazon Ring's New Indoor Surveillance Drone Is Not Compatible with Privacy (November 16, 2026). https://www.eff.org/deeplinks/2026/11/amazon-rings-new-indoor-surveillance-drone-not-compatible-privacy
  11. Ars Technica, Jon Brodkin - Prisons get OK to jam contraband phones. CTIA says nearby phones will be jammed too. (November 20, 2026). https://arstechnica.com/tech-policy/2026/11/prisons-get-ok-to-jam-contraband-phones-ctia-says-nearby-phones-will-be-jammed-too/
  12. EFF Deeplinks, Beryl Lipton - Voters Reject ALPRs and Surveillance Tech at the Ballot Box (November 18, 2026). https://www.eff.org/deeplinks/2026/11/voters-reject-alprs-and-surveillance-tech-ballot-box
  13. 404 Media, Matthew Gault - Black Neighborhoods Are Twice as Likely To House a Data Center, Study Finds (November 19, 2026). https://www.404media.co/redlining-ai-data-centers-and-race-study/
  14. Ars Technica, Dan Goodin - Man hired by Russia spied on Silicon Valley tech and defense company, feds say (November 19, 2026). https://arstechnica.com/security/2026/11/man-charged-with-spying-on-silicon-valley-tech-and-defense-company-for-russia/