> daily_signal(2026_06_16)

The law that lets the government surveil without a warrant lapsed, but the surveillance kept running. The DOJ called Musk's unpermitted gas turbines a national-security matter.

PickBits Daily Signal · Tuesday, June 16, 2026

By Mark Pickering · 11 min read · June 16, 2026

// tl;dr

FISA Section 702, the law that lets the government collect Americans' international communications without a warrant, lapsed at midnight on June 12 for the first time since 2008. The House rejected a short-term extension 218 to 198; under fast-track rules it needed a two-thirds majority and did not even reach a simple one. The fight was over Trump naming Bill Pulte acting intelligence chief. But the surveillance court already renewed its collection orders back in March, and those run through March 2027, so the program keeps operating while the law that authorizes it is dead.
The Justice Department asked a court to throw out an NAACP lawsuit over xAI's 57 unpermitted gas turbines near Memphis, arguing that switching them off would threaten "national, economic, and energy security." A Defense filing said Grok's government model is one of four AI systems supporting "mission-critical operations," including recent US strikes against Iran. The turbines power Musk's Colossus AI plant in a neighborhood that is already among the most polluted in the country.
For the first time, the White House used national-security export powers to force a US company to pull an AI model, giving Anthropic 90 minutes to restrict two of its most advanced cyber models. The stated reason was a jailbreak of one model that Amazon researchers found. But Semafor and Fortune report the real trigger was a feared China-linked breach of the other, Mythos. Anthropic says the government never raised China with it. The company pulled both models from the market entirely.
A man with ALS has now spoken through a brain implant for more than 3,800 hours at home, the first long-term proof that a speech brain-computer interface works for daily life outside a lab. Over 22.6 months, Casey Harrell used his implant to talk with his family, read to his young daughter, and do his job, at up to 99% accuracy across a 125,000-word vocabulary. The results were published in Nature Medicine.

Ten days ago the story here was the government drawing lines around AI to protect people: a state legislature voting to freeze new data centers, the Supreme Court letting fines stand against carriers that sold your location, a court agreeing to weigh whether the maker of an AI tool can be sued. This week the arrow points the other way. Three of today's four stories are the same shape, and the shape is a phrase: national security. It is the argument that kept the wiretaps running after the law authorizing them expired. It is the argument the Justice Department used to shield an unpermitted power plant from the people breathing its exhaust. It is the argument the White House used to pull an AI model off the market with 90 minutes' notice. The phrase is doing a lot of work right now, and it is worth watching which doors it opens. The fourth story is the one that needs no such justification, because all it does is give one person his voice back.

This week the warrantless-surveillance law lapsed while the surveillance kept running, the DOJ called Musk's unpermitted turbines a national-security asset, the White House forced an AI company to pull two models in 90 minutes, and a man with ALS passed 3,800 hours of speaking through a brain implant at home.

1. The warrantless-surveillance law expired for the first time since 2008, and the surveillance did not stop.

The clearest case this year of a check expiring while the power it was supposed to limit keeps running.

At midnight on June 12, FISA Section 702 lapsed. It is the law that lets US intelligence agencies collect the phone calls, texts, and emails of foreign targets without a warrant, a net that routinely sweeps up the communications of Americans those targets talk to. It had been renewed every few years since 2008 with little drama. This time the House rejected a short-term extension to July 2 on a bipartisan vote of 218 to 198; because leadership ran it under expedited rules that require a two-thirds majority, the measure failed without even reaching a simple one. Intelligence officials say more than 60% of the material in the president's daily brief comes from collection under this authority, which is the number they lead with whenever it is up for renewal.

What broke the renewal was not a sudden privacy awakening. It was a fight over Bill Pulte, whom President Trump installed as acting Director of National Intelligence, a job Pulte has no national-security background for. Democrats and at least 19 House Republicans balked at extending expanded surveillance tools to an office in his hands. Here is the part that matters for everyone: the lapse does not turn the program off. The Foreign Intelligence Surveillance Court renewed its collection certifications back in March, and those orders run through March 2027. So the statute that authorizes the program is dead, but the court orders that actually run it are in force for roughly nine more months. The law lapsed; the wiretaps did not.

NPR report June 12 2026 FISA Section 702 warrantless surveillance law lapses for the first time since 2008 House rejects extension Pulte acting DNI standoff foreign intelligence surveillance court certifications run through March 2027 — npr.org · June 12, 2026

Why this matters: If you text or call anyone outside the country, or anyone who does, your communications have always been collectible under this program without a warrant, and the expiration of the law does not change that, because the court orders that run the collection outlive the statute by nine months. The headline says a surveillance power expired; the reality is that the thing that expired was the requirement for Congress to vote on it, not the surveillance. That is the gap to hold onto: a power can outlast the law that created it when the machinery is already running. Action this week: If you want your everyday messaging outside the reach of this kind of bulk collection, move it to an end-to-end encrypted app like Signal (signal.org) and turn on disappearing messages, which takes about ten minutes and means the content is unreadable in transit. The Electronic Frontier Foundation's plain-English explainer of what 702 does and does not cover is at eff.org/702, worth ten minutes if you have ever wondered what is actually collected.

npr.org: FISA Section 702 surveillance law expires amid the Pulte standoff (June 12, 2026)
thehill.com: FISA 702 spy powers set to expire after House vote fails over Pulte backlash (June 12, 2026)

2. The Justice Department told a court that shutting off Musk's unpermitted gas turbines is a national-security risk.

The data-center power fight that ran through coal plants and the New York moratorium this month now has the federal government on the company's side.

In a court filing dated June 16, the Justice Department asked a judge to dismiss an NAACP lawsuit aimed at xAI's gas turbines outside Memphis, arguing that stopping them would "threaten American national, economic, and energy security" and undermine "artificial-intelligence innovation that supports the Department of War's" military operations. A Defense filing in the case stated that Grok's government model is one of four AI systems supporting "mission-critical operations," including the recent US strikes against Iran. The turbines power xAI's Colossus supercomputing site, and the number on the ground has grown to 57, more than double an earlier count, with Musk's company planning another $2.8 billion in turbines over three years.

The NAACP, represented by the Southern Environmental Law Center, filed suit in April. Its core claim is that the turbines are unpermitted: they sit on trailers, which the lawsuit argues is a way to class them as temporary and dodge air-pollution permitting, in a part of south Memphis that is already among the most polluted in the country. The pollutants the plant adds, including nitrogen oxides, fine particulate matter, and formaldehyde, are the ones tied to asthma and heart disease. Strip away the framing and the filing is a remarkable move: the federal government is arguing that because an AI chatbot has been folded into military operations, the unpermitted power plant feeding it should be insulated from the environmental law the people next door are trying to enforce.

TechCrunch report June 16 2026 DOJ claims xAI unpermitted gas turbines are a matter of national economic and energy security NAACP lawsuit Memphis Colossus Grok Department of War Iran strikes Southern Environmental Law Center — techcrunch.com · June 16, 2026

Why this matters: If a data center or one of these "mobile" trailer-mounted power plants is ever proposed near you, the Memphis case is now the template for the strongest argument a company can make to keep it running: that the AI it powers is wired into national defense, so an environmental challenge has to give way. The people living next to xAI's site are breathing the exhaust today, and the government's response to their lawsuit is not that the pollution is acceptable, but that the plant is too important to national security to stop. That is a precedent worth knowing before it shows up in your county. Action this week: Check the air quality where you live at airnow.gov, which takes about two minutes and tells you what you are actually breathing. If a data center or a trailer-mounted gas plant is proposed near you, look up whether it has an air-pollution permit on file with your state environmental agency before the first hearing, because "it is temporary" is the move the Memphis turbines used to avoid one.

techcrunch.com: DOJ claims xAI's unpermitted gas turbines are a matter of national, economic, and energy security (June 16, 2026)
actionnews5.com: Elon Musk's xAI company seeks dismissal of NAACP lawsuit (June 16, 2026)
gizmodo.com: Under Trump, DOJ moves to intervene in NAACP lawsuit supporting Musk's xAI (June 16, 2026)

PickBits Daily Signal is free. If this lands in your inbox every day and it is worth something to you, the best way to support it is to share it with someone who would read it. Subscribe at pickbitsai.substack.com.

3. The White House forced Anthropic to pull two AI models in 90 minutes, the first time export power has been used on AI software.

The novel part is not that the government acted; it is the legal tool it reached for, and the reason it gave does not match the reason that has been reported.

On June 12, the White House gave Anthropic 90 minutes to restrict access to two of its most advanced cybersecurity-focused models, Claude Fable 5 and Mythos 5, ordering the company to limit them to US nationals and roll back global access. Rather than comply piecemeal, Anthropic pulled both models off the market entirely. What makes this more than a one-off dispute is the authority the government used: by Anthropic's account the directive cited national-security export-control powers, and it appears to be the first time those powers have been used on AI software rather than chips or hardware. The same authority that controls which countries can buy advanced processors was just pointed at a piece of software, which means a model that hundreds of millions of people use can be ordered offline the way a shipment of chips can be blocked.

The reason given does not sit cleanly. The official basis, Anthropic says, was a jailbreak of Fable 5, a method Amazon researchers found to coax restricted cyberattack information out of the model, and Anthropic disputes that it warranted a recall: "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people." But Semafor and Fortune report that the real trigger was something else: a suspicion that a China-linked group had accessed Mythos, raising the fear that Beijing could copy the model through "distillation," studying its outputs to rebuild its capabilities. Here is the catch worth flagging: an Anthropic spokesperson said the White House never raised Chinese access to Mythos in its conversations about the jailbreak and the export order. So the public reason and the reported reason are two different stories, and the company at the center of it says the two were never connected in the room. Pete Hegseth had already designated Anthropic a "supply chain risk" in February.

The Record report June 15 2026 Anthropic says government forced it to disable cyber AI models Fable 5 Mythos 5 national security export control first time used on AI software not hardware 90 minutes jailbreak — therecord.media · June 15, 2026

Why this matters: If you or your team build anything on top of a single company's AI model, this is the week to notice that a model you depend on can be ordered offline by the government with 90 minutes' warning, and the company building it may not even get a reason that matches what is actually driving the decision. The deeper issue is that "national security" was enough to pull a product used by hundreds of millions of people, on a legal authority that had never been pointed at software before, with the stated reason and the reported reason pointing in different directions. When the justification and the action do not line up, the precedent is what survives. Action this week: If your work or your product hard-wires one proprietary AI model, spend 20 minutes writing down which of your tools would simply break if that model vanished tomorrow and which have a fallback you could switch to. That list is now a real continuity risk, not a hypothetical, and the cheapest time to find the single points of failure is before one of them disappears on an export order.

therecord.media: Anthropic says government forced it to disable cyber AI models (June 15, 2026)
fortune.com: How a warning from Amazon led the White House to shut down Anthropic's Mythos model (June 14, 2026)
semafor.com: White House move to limit Anthropic linked to concerns about Chinese access to Mythos (June 13, 2026)

4. A man with ALS has spoken through a brain implant for 3,800 hours at home.

The constructive story of the week, and the one piece of AI here that needed no national-security argument to justify itself.

Researchers at UC Davis reported that Casey Harrell, a 45-year-old man paralyzed by ALS, has become the first long-term "power user" of a speech brain-computer interface. Four arrays of electrodes were implanted in his brain in a five-hour surgery in July 2023; the device reads the neural signals he would use to speak and turns them into words on a screen, in close to real time. The new finding, published in Nature Medicine, is not the implant itself but the endurance test: over 22.6 months, Harrell used it for more than 3,800 hours at home, with no researchers present. He uses it to talk with friends and family, to read to his young daughter, and to do his job as an environmental activist.

The accuracy is the part that moves this out of the demo category. On day one the system hit 99.6% on a 50-word vocabulary; expanded to 125,000 words it held 97.5%, and the current version runs near 99%. "He's the first power user of a speech BCI," said UC Davis neuroengineer Sergey Stavisky, whose colleague David Brandman led the surgical team. Harrell's own summary is the line worth keeping: "Any one of these things would be an absolute godsend of improvement. To have all of them is truly revolutionary." This is still a single patient, and a brain implant is not a consumer product. But after years of one-off lab demonstrations, this is the first evidence that the technology holds up across nearly two years of ordinary daily use, which is the only test that ever mattered for the people who need it.

MIT Technology Review report June 15 2026 Casey Harrell ALS first power user of speech brain implant 3800 hours at home 22.6 months UC Davis Sergey Stavisky David Brandman Nature Medicine 99 percent accuracy 125000 word vocabulary — technologyreview.com · June 15, 2026

Why this matters: If you have ever doubted that any of this AI delivers something that lands squarely on the side of a person rather than a company or a government, here is the clean answer: a man who lost his voice to ALS has spent nearly two years talking to his family and reading to his daughter through a brain implant, at home, at 99% accuracy. There is no national-security argument attached, no surveillance angle, no liability fight, just a capability that gives one person back the thing the disease took. It is also a fair reminder of what real progress looks like next to hype: not a launch event, but 3,800 hours of a tool working in someone's living room. Action this week: Read the MIT Technology Review piece (technologyreview.com, "The first brain-implant power user") for one concrete, non-hype example of AI doing good to point to the next time the conversation is all doom. If someone in your life has ALS or another condition that takes speech, the programs to watch are BrainGate2 and the UC Davis neuroprosthetics trials, both searchable on clinicaltrials.gov.

technologyreview.com: A man with ALS is the first power user of a speech brain implant (June 15, 2026)
nature.com: Nature Medicine, long-term home use of a speech neuroprosthesis (June 2026)

» What to watch this week

Whether Congress writes a new FISA Section 702 with reforms, or lets the court certifications quietly carry the program to March 2027. The cleanest tell of whether the lapse means anything is whether anyone moves to attach warrant requirements before the next vote, or whether the lapse just becomes a year of collection with no statute behind it.
Whether the judge in the Memphis case accepts the national-security argument as a reason to dismiss the NAACP suit. If "this plant powers a model the military uses" is enough to end an environmental case, that reasoning will be in every data-center fight by fall. Watch for the ruling and for the first developer elsewhere to borrow the language.
Whether Anthropic gets a written, specific reason for the export order, and whether the China-access reporting is ever confirmed or denied on the record. The gap between the stated jailbreak rationale and the reported Chinese-breach trigger is the whole story; an official accounting either closes it or makes the precedent more alarming.
Whether the UC Davis team moves toward a version that does not require an open skull connection. The 3,800-hour result proves daily use works; the next milestone that decides whether this reaches more than a handful of patients is a fully implanted, wireless system, so watch for that announcement.

Tomorrow's signal lands here.