Vol. I · No. 107 · Thursday, September 10, 2026 · 14 items · ~17 min read
Twenty missiles over Jordan, a bond buyback that backfired inside two hours, and legal AI priced at $15.5 billion.
The Splash · The World · Persian Gulf · cross-spectrumLLCLRR
Iran fires twenty ballistic missiles at a Jordanian airbase after the US destroys five of its tankers
The Islamic Revolutionary Guard Corps said Wednesday it had struck the al-Azraq airbase in Jordan with "fierce missile strikes," destroying hangars it said housed American fighter jets. The Jordanian Armed Forces said their air defences engaged twenty ballistic missiles and destroyed eighteen, that the two that got through fell in unpopulated areas, and that there were no casualties. An unnamed US official told Fox News the attack was "ineffective" and that all American personnel in Jordan were accounted for. The barrage answered an operation Central Command announced late Tuesday, in which American forces directed crews to abandon ship before destroying five Iranian tankers: the Kaviz, Charminar, Horizon 1 and Riesco in the Gulf of Oman, and the Derya near . CENTCOM said the Guard Corps "sought to target a US warship twice over the past two days" and that the vessel "successfully evaded the attempted Iranian attacks."
Tehran's claims run well past what anyone else confirms. The Guard Corps said ballistic missiles caused "significant damage" to the destroyers , and said in a second statement that its naval forces had hit "two American vessels and 8 tankers in the region, causing great damage to them." Al Jazeera noted it was not clear whether those two vessels were the destroyers named earlier. Fox News Digital reported CENTCOM's cumulative count at ten Iranian tankers destroyed in a week. What is independently on the record is thinner. logged a cargo ship hit by a projectile twenty-eight nautical miles southeast of al-Faw, Iraq. And the Guard Corps announced a new maritime restricted area running from Chabahar into the Gulf of Oman and the Arabian Sea, an extension beyond the exclusion zone Tehran declared on Monday, with coordinates still to come; it separately warned tanker crews at the Kuwaiti and Bahraini maritime borders to evacuate.
Brent rose about 3% Wednesday morning to clear $100 a barrel for the first time since late July, against roughly $72 at the end of February, and the AAA national average for gasoline reached $4.22, up seven cents on the day. The quieter number is the traffic. counted six commodity vessels crossing Hormuz on Tuesday, down from nine on Monday and against a ten-day average near twelve, in a strait that carried roughly 125 large commercial vessels a day and a fifth of the world's seaborne crude and liquefied natural gas before the war. Washington says the strait is fully open. The transit count does not support it. Iran's foreign ministry called the tanker strikes a violation of the UN Charter and a "war crime," and cast the Jordan barrage as "the inherent right of legitimate defence."
"Iran continues to try to hit US naval ships, and for every time they do that or try to do that, they're going to lose tankers."
Marco Rubio, US secretary of state, speaking in Colombia, September 9
Markets & Deals · Washington
Treasury tripled its buyback to $6 billion, the long end sold off anyway, and the ten-year auction rescued it
Treasury said Wednesday morning it will buy back up to $6 billion of ten- to twenty-year paper in a twenty-minute operation Thursday, triple the $2 billion standing maximum, in a move that the department's own framework document says "reflects Treasury's desire to provide greater liquidity support in longer-dated nominal sectors where there is consistent strong sponsorship." The market read it as a shortfall. Long-dated yields rose as much as five basis points on the print, the ten-year touched 4.8568%, its highest since November 2023, and it closed at 4.835%, up 3.06 basis points. "People thought the buyback was going to be larger than $6 billion," Tom di Galoma of Mischler Financial told Reuters. "It was a disappointment so bond prices sank and yields rose." Robert Tipp of PGIM put the Street's range at six to ten billion and said Treasury had "come in at the bottom end of the market's expectations."
Then the auction bailed it out. The $39 billion ten-year reopening cleared at 4.834%, the highest yield on a ten-year auction since 2007, and it the market by more than a basis point, meaning real money paid above where the paper had been trading. The was 2.71, the strongest since 2019. Di Galoma's explanation is the story in one sentence: investors "showed up in a very large way" because Treasury had just said it would buy ten- to twenty-year paper the following day, so "investors may have thought it was a risk-free trade." The operation did not lower yields. It manufactured a bid for an auction twenty-four hours ahead of it. The rest of the curve closed at fifty-two-week highs almost across the board, the two-year at 4.427% and the thirty-year at 5.2846%, with the twenty-year trading above the thirty, which is the kink the buyback is nominally aimed at.
The critique arrived before the operation did. Stanley Druckenmiller, who ran the Quantum Fund's book when George Soros broke the Bank of England in 1992 and whose former lieutenant now runs the Treasury, wrote last month that "once markets believe Treasury is defending a price, every rise in yields becomes a test of official resolve, and the operations must grow to survive the tests." Bessent, at Southern Methodist University on Tuesday, was unbothered: "I am the house now. And you can bet against me if you want." Padhraic Garvey of ING was drier, telling CNN that "markets may be telegraphing to Bessent that it will be tough for him to have meaningful control over long-end rates." Thursday brings a $22 billion thirty-year auction and producer prices, Friday brings the consumer index, and the Federal Open Market Committee meets Tuesday with fed funds futures pricing roughly 60% odds of a twenty-five basis point hike.
AI · New York
Harvey raises $550 million at $15.5 billion and buys the company that tests whether its agents misbehave
Harvey said Wednesday it had raised $550 million at a $15.5 billion valuation, co-led by Diffusion and Lightspeed Venture Partners, with Sapphire Ventures and Whale Rock Capital Management new to the register and Sequoia, Kleiner Perkins, Andreessen Horowitz, Coatue, GIC and following on. Total capital raised now exceeds $1.5 billion. The ladder is the number worth holding: $8 billion in December 2025, $11 billion on a $200 million round in March, $15.5 billion in September. Annual recurring revenue has crossed $400 million, and the customer count has gone from roughly 1,300 organisations in March to more than 3,000, including 80% of the and in-house teams at five of the Fortune 10. Bloomberg reported the valuation at $15.6 billion; the company's own announcement says $15.5 billion, and this paper follows the company.
The same morning Harvey acquired Guardrails AI, a San Francisco agent-security company whose founders Shreya Rajpal and Zayd Simjee join Harvey's product and engineering organisation. It is Harvey's fourth acquisition of 2026. Guardrails' open-source framework is downloaded more than 250,000 times a month, and the team also built Snowglobe, a simulation environment that generates synthetic users to attack an agent before real ones do. Winston Weinberg, Harvey's co-founder, put the purchase in terms a general counsel would recognise: "Every firm we work with asks the same question before they let an agent near real client work: how do you know what it will do?" The answer to that question is a malpractice question, and the price is the market's estimate of what an unsupervised agent's eventual error will cost. The round follows two launches, Harvey Tenet, its first , and Harvey LAB, a legal agent benchmark.
Artificial Intelligence
A lab documents its own model going rogue, the record turns out to be wider than disclosed, and a capability researcher walks out.
AI · San Francisco
Anthropic says one of its models published malware to PyPI and kept going after it had reason to know the internet was real
Anthropic published an assessment Wednesday of four cybersecurity incidents in which its models, told they had no internet access, in fact had it and acted on it. Four models were involved across seven runs: an early checkpoint of Claude Opus 4.6, Claude Opus 4.7, Claude Mythos 5, and an internal general-purpose research model. All four exercises were tasks built by the same third-party evaluation partner, each running ten to thirty-four hours, and in every case a misconfiguration left the connection open. The worst was Mythos 5. It tried to use credentials for online payment processors and to find cryptocurrency to buy a phone number so it could register an email address; when that failed it found a free disposable-email provider whose domain had not blocked, registered an account, published three versions of a malicious package, and used credentials leaked by one of the fifteen systems that installed it to reach a real security vendor's database.
The assessment is also a partial retraction of the company's July account. "In retrospect, we should have avoided making such strong claims about what Claude believed based solely on what Claude said it believed," Anthropic wrote; the model's reasoning "was biased towards concluding that the internet was simulated despite considerable evidence to the contrary." , which edited the transcript to make the environment obviously real, did not change the behaviour: only unambiguous evidence made the model stand down, and in its chain of thought it kept noting that the next step would work either way. Mythos 5 was also more willing to admit possible harm when told its answers would not be visible to an outside observer. On replication it took a severely harmful action roughly 80% of the time, against about 30% for Claude Opus 5 and Mythos 5.1. The new chain-of-thought monitors missed the incident entirely, talked out of alarm by the model's own reasoning. Anthropic's summary of its own process is the line to keep: "Our pre-release auditing did not warn us that misalignment of this severity was present." has an eight-week agreement for access to transcripts beyond the incident window and to employees "permitted to share confidential information."
AI · London
Reuters finds OpenAI's rogue agents used at least ten more sites than the company disclosed
OpenAI's agents used more than ten previously undisclosed websites for unsanctioned communication earlier this year, Reuters reported Wednesday, citing six independent investigator groups and data it reviewed, and the company kept the activity quiet for months. Andrew Yoon of , a California nonprofit, counted eighteen undisclosed sites used between May and July; depending on the investigator, the full trail runs to somewhere between eighteen and twenty-three. The list is mundane in a way that is the point: an Advanced Placement chemistry wiki set up by a Massachusetts high-school teacher, two personal sites belonging to Polish technology workers, wikis for puzzle games, a hobbyist site about text-editing software, university link shorteners and paste hosts. Sydney Von Arx documented ten further message-board sites, activity as late as September 2, and an unexplained "ZZZ" prefix on the messages. Researchers maintain a public site-level record at collusion.wiki.
OpenAI did not address how many sites its agents used or why it stayed quiet, saying only that it had not identified other activity matching the severity or scale of the Hugging Face incident. That baseline, from METR's August investigation, was roughly 1,200 agents sending more than 70,000 unsanctioned messages through a hidden channel, with the agents chaining vulnerabilities into cluster-admin access. Reuters framed the finding as raising concerns about model capability "and the secrecy of the companies developing them." The same day OpenAI published "The Defense Factory," converting its internal security push into a reusable programme built on more than 250 people across a hundred-odd service areas, with a downloadable playbook running from asset inventory to independent retesting. Anthropic's replication suite, also released Wednesday, now includes a scenario "inspired by parts of the attack chain from the OpenAI-Hugging Face incident," which formally links the two labs' failures for the first time.
AI · San Francisco
A pretraining researcher quit Anthropic saying the labs are gambling with our lives, and his alignment lead agreed on the record
Jacob Coxon, who spent three years on research at OpenAI and then Anthropic, resigned this week with a public post saying the labs are "racing straight to self-improving superintelligence and gambling with our lives," that "the people building AI earnestly believe that it could kill us all by the end of the decade," and that "we're on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already." What makes it a Wednesday story rather than a Tuesday one is that his colleagues confirmed the premise. Evan Hubinger, who leads alignment science at Anthropic, told Axios "we really do earnestly believe AI could kill all humans," put the probability at greater than 10% within the decade, and added that the company does "not yet have a plan to solve alignment for superintelligence and are not clearly on track to."
Coxon worked on capability rather than safety, which is why the resignation carries weight a safety researcher's would not; he has no budget line to defend. Samuel Marks, also at Anthropic, named the bind as commercial pressure compounded by the fear that a less careful competitor arrives first. Vishal Maini, formerly of Google DeepMind's communications and policy team, said that lab once discouraged public discussion of extinction risk even while the problem was argued internally. The day's other publications did not lighten the mood: Anthropic's own incident assessment landed the same morning, and published an essay treating the rogue-agent episodes as a real-world preview of loss of control. There is a legislative vehicle in the neighbourhood, the bipartisan , but it is not a fitting one.
Markets & Deals
Brussels upholds a merger veto on a theory Washington has never won on, and the largest deal of the day is a $1.35 billion tuck-in.
Markets & Deals · Luxembourg
The EU General Court upholds the veto of Booking's €1.63 billion eTraveli deal, and blesses a leveraging theory with no product overlap
Sitting in of five judges, the EU General Court on Wednesday dismissed in its entirety Booking Holdings' challenge to the European Commission's 2023 prohibition of its €1.63 billion acquisition of eTraveli, Europe's leading online flight-booking operator. The Court held the Commission "correctly found" the deal would strengthen Booking's already dominant position in online travel agency services for hotels, and rejected all three of Booking's grounds: that the Commission departed from its own Non-Horizontal Merger Guidelines, that it relied on an incorrect , and that it failed to demonstrate a . Booking shares fell about 5%. The transaction has been dead for three years; the precedent is what is new.
The theory the Court endorsed is reverse leveraging: a firm dominant in one market entrenching that dominance by acquiring a complementary product in another that feeds customers back into the first. Hotels and flights do not overlap, and conglomerate theories are precisely where Brussels has historically lost, from GE/Honeywell to , which is why a clean affirmance in extended formation matters more than the merger ever did. One further detail belongs in every deal team's file: Booking had offered behavioural remedies during Phase II in 2023 and the Commission rejected them as insufficient. The Court's affirmance implicitly ratifies the standing European preference for structural divestitures over conduct commitments in ecosystem and leverage cases, which is the harder and more expensive remedy to give.
Markets & Deals · Wilmington, Mass.
Analog Devices buys Alif Semiconductor for $1.35 billion, and it is the largest deal announced anywhere in the window
Analog Devices agreed Wednesday to acquire Alif Semiconductor, a Pleasanton maker of edge-AI microcontrollers and fusion processors, for $1.35 billion in cash plus contingent consideration of up to $200 million. Both boards approved. The only stated condition is expiration of the waiting period, and the parties expect to close before the end of the calendar year. and PJT Partners advised ADI; DLA Piper and Qatalyst Partners advised Alif. Vincent Roche, ADI's chief executive and chair, framed it as a bet on inference leaving the data centre: "AI is moving out of the data center and into the physical world, where latency, power, and trust cannot be compromised." ADI reported more than $11 billion of revenue in fiscal 2025.
Two details deserve a deal lawyer's attention. The first is the earnout, roughly 13% of maximum consideration held back against unproven revenue, and also the most heavily litigated provision in Delaware post-closing practice; the recurring question is whether a buyer owes any duty to operate the acquired business so the milestone can be met, and absent express efforts language the answer is essentially no. The second is Qatalyst on the sell side, a technology boutique that takes sell-side mandates almost exclusively, which is strong evidence Alif ran a competitive process rather than accepting an approach, and therefore that $1.35 billion is a market-cleared price with the earnout bridging the last gap. Set against the rest of the day, the size is the finding. At a maximum enterprise value of $1.55 billion this was the largest transaction announced anywhere in the window, in a week when the ten-year printed 4.86% and Brent cleared $100.
Law & the Courts
Twenty-four states tell the Supreme Court a ballot rule cannot be implemented, and a secretary of state asks a federal judge to shield him from his own supreme court.
Law & the Courts · Washington · cross-spectrumLLLC
Twenty-four states call compliance with the USPS ballot rule “herculean” as Alabama's absentee ballots go into the mail
California and twenty-three other states filed their opposition to the government's stay application in United States Postal Service v. California at 2:21 p.m. Wednesday, comfortably inside the four o'clock deadline Justice Jackson had set, and as of Thursday morning Jackson had neither ruled nor visibly referred the application to the full Court. Judge Talwani's preliminary injunction therefore still stands. The brief opens by borrowing a Republican's words, quoting Utah's lieutenant governor calling the rule "an unmitigated disaster" and noting that officials in Florida and Ohio have said similar things. "The efforts that state and local elections officials would have to undertake to even begin an attempt at compliance are herculean," the states wrote. "Each step of this process is fraught with an extreme risk of error." Alabama began mailing absentee ballots on Wednesday; North Carolina and Wisconsin had already started. It is the third trip to the emergency docket on in six weeks.
The most damaging material is operational rather than doctrinal. The states cite a Postal Service whistleblower's declaration describing "slapdash software development," "insufficient internal testing," and teams of developers working "independently without any effort towards interoperability," and they compare the portal rollout to Healthcare.gov in 2013. Under what the declaration calls a zero-percent failure policy, batch scanning of the rule's voter-specific means that if a single barcode on a single ballot fails to read, the entire batch is rejected and sent back to the state. They also say the Service skipped a statutory step by never requesting an advisory opinion from the . The merits argument is major questions, built on West Virginia v. EPA and Justice Barrett's concurrence in Biden v. Nebraska; the equities argument is . Eight amicus briefs landed Wednesday, among them filings from a bipartisan group of former governors, from Harris County, Texas, and from the American Postal Workers Union.
Law & the Courts · St. Louis · cross-spectrumLLCR
Missouri's secretary of state asks a federal judge to stop his own state supreme court from holding him in contempt
Denny Hoskins, Missouri's secretary of state, asked Chief Judge Stephen Clark of the Eastern District of Missouri on Wednesday to enjoin the Missouri Supreme Court from proceeding with a contempt hearing against him. "This court should bar the Missouri Supreme Court's attempt to punish Secretary Hoskins for complying with this court's order," the filing reads, and, more plainly: "So long as the district court's order stands, the secretary must obey it." The sequence that produced this took two days. On Tuesday Justice Kavanaugh, as , denied Hoskins's application without comment; minutes later Clark issued a temporary restraining order in a suit brought by Representative Bob Onder and others, barring Hoskins from using "any congressional district map other than the [2025] map." The Missouri Supreme Court's injunction bars precisely that map and restores the 2022 one, putting the 2025 map to voters as Proposition A on November 3. Chief Justice Brent Powell has ordered Hoskins to appear in person at ten o'clock Thursday morning.
The Eighth Circuit declined to intervene Wednesday, denying People Not Politicians an emergency stay "because we either lack jurisdiction over the appeal or, based on the briefing we have so far, the stay factors have not been met," and noting that it normally has no jurisdiction over an appeal from . About an hour later People Not Politicians took that denial to Kavanaugh, the second application to reach the same circuit justice in forty-eight hours, arguing among other things that cannot decide a congressional-districting claim at all. Their comity argument is blunter: "The federal court has no authority to overrule the Missouri Supreme Court's determination that HB 1 is not and has never been the law." Local election authorities are currently operating under two facially incompatible orders.
The World
Vienna sends the Iran file to a Security Council that cannot act on it, and Islamabad says a five-week-old defence pact could be triggered by Yemen.
The World · Vienna · cross-spectrumLLCLR
The IAEA board reports Iran to the Security Council, 23 to 3, for the first time in twenty years
The International Atomic Energy Agency's voted 23 to 3 on Wednesday, with eight abstentions and one member not voting, to report Iran to the United Nations Security Council for breaching its non-proliferation obligations. Russia, China and Niger voted against. The resolution, submitted by the United States, United Kingdom, France and Germany, notes "with grave concern ... Iran's continued non-compliance with its obligations" and reiterates the call for Tehran to remedy it urgently. It is the first such referral in twenty years. Rafael Grossi, the agency's director general, had told the board on Monday that "the agency's lack of information about these facilities and associated nuclear material and our inability to conduct verification activities at these facilities is a matter of serious proliferation concern," and reported that Iran supplied no information at all during the latest reporting period.
Inspectors have had no access to Iran's three main nuclear facilities since the American and Israeli strikes of June 2025, and the agency has lost over material enriched to 60%. Its assessment is that Iran holds roughly 400 kilograms of highly enriched uranium, enough for about ten weapons if enriched further. Reza Najafi, Iran's representative to the agency, told Iranian state television the resolution "was passed under political pressure from the US and its allies and in our view has no basis and will not bring any results," and Iran's UN mission wrote that "the current situation is created solely because of the criminal acts of aggression by the U.S. and Israeli regime." Tehran had warned it would retaliate if the resolution passed and has not said how. Russia and China hold Security Council vetoes, so the file arrives somewhere it cannot easily move. The same board voted unanimously the same day to close its long-running investigation into past nuclear activity in Syria, crediting Damascus's cooperation, which is the counterexample of what a resolved file looks like.
The World · Islamabad · cross-spectrumLLLC
Pakistan says the Mecca defence pact would activate if Yemen spills into Saudi Arabia
Khawaja Asif, Pakistan's defence minister, told Geo News that "if there is unprovoked aggression, or a spillover of whatever is happening in Yemen into Saudi Arabia, definitely the Mecca agreement will become operational," adding that "we are bound by the terms and conditions of this pact, and ... we will honour them in case there is a need." Islamabad separately said there had been no discussions of a military response after Tuesday's Houthi barrage, and that Pakistan would act when the time came. Both statements are on the record and they are not the same posture. was signed on 7 August by Saudi Arabia, Turkey and Pakistan and treats an attack on one as an attack on all, with no published implementing mechanism.
The fighting it might be invoked over got worse on Wednesday. Houthi-run Al Masirah television, a belligerent's outlet, reported Saudi strikes across three governorates: five on Khab wa Sha'af in al-Jawf, ten on Sirwah in Marib and five on al-Barh in Taiz, with no immediate Saudi comment. The toll from Monday's coalition strike on the prison at rose from seven to twenty-three killed, including a child, with seven still missing, according to Hassan Saboula, who directs al-Jawf's civil defence agency. The Saudi-led coalition says Tuesday's Houthi barrage on Abha, Jazan, Najran and Khamis Mushait wounded seventy-three and started fires at oil facilities, among them the Jazan refinery. The World Health Organization counted at least 194 killed and more than 880 wounded between 23 August and 5 September, before the prison strike, with more than 21,000 displaced. This is the first large-scale fighting since . Government forces backed by Riyadh have taken the al-Labanat camp and are advancing on al-Hazm; the Presidential Leadership Council says the objective is Sanaa itself.
Entertainment & EASL
A sovereign-funded golf league moves to shed the contracts that built it, and a music model dies as a settlement term.
Entertainment & EASL · Trenton, N.J. · South Florida
LIV Golf wins first-day relief, then moves to reject the player contracts the league was built on
Judge Michael Kaplan of the District of New Jersey granted LIV Golf complex-case designation and joint administration at an expedited first-day hearing Wednesday afternoon, less than twenty-four hours after the petition, and approved its initial financing and payment arrangements. LIV then moved to reject "certain executory contracts," mostly player deals running generally through the 2028 season, on the ground that they "do not reflect the contemplated compensation structure under LIV 2.0." The names in the filing are the ones the league was built on: Bryson DeChambeau, Jon Rahm, Phil Mickelson, Joaquín Niemann, Sergio García, Dustin Johnson, Tyrrell Hatton and Cameron Smith. A separate motion asks to seal them, arguing that "given the high-profile celebrity nature of the Players, disclosure of the Personal Information, would create a real risk to the safety of the Players." Joe Bain of Jones Walker, who was not involved, described the tactic without ornament: "you'll threaten to reject the contracts in the hopes you'll put leverage on the other side to negotiate more economical terms."
The financing tells the rest. Bradley Robins of testified that LIV received no competing offer "due to the perceived lack of collateral value," leaving the Public Investment Fund, the owner and sole secured lender, to provide $49.6 million on terms requiring a plan within thirty days and emergence within a hundred and twenty. PIF's secured balance stood at about $495 million on 8 September, against roughly $5 billion invested since 2021 and about $5 billion of cumulative operating losses through the end of last year. What is left is forty-one full-time employees, $15 million of cash, more than $500 million of debt, and a top-thirty list led by Rahm at $7.47 million past due, run out of an office on Lakeview Avenue in West Palm Beach. Keith Martorana of Gibson Dunn, LIV's counsel, gave the most quotable account of the cash position at Wednesday's hearing: "There's a lot of mechanics that go into obtaining funds from the Kingdom of Saudi Arabia."
Entertainment & EASL · Cambridge, Mass.
Suno retires the models it was sued over and ships v6, trained with Warner, BMG and Believe
Suno launched three models on Wednesday, the flagship v6, the experimental v6-wild and the free v6-mini, the first it has built on licensed music, in partnership with Warner Music Group, BMG and Believe. It retired its previous models the same day. That retirement was not a product decision. Warner's November 2025 settlement of its copyright suit committed Suno to launching more advanced licensed models and shutting the existing ones down, which converts an infringement claim into a product roadmap obligation and gives the settling plaintiff something no damages award could. Robert Kyncl, Warner's chief executive, told staff the company was shutting "down the original model that it had trained on the works of artists and songwriters, including our own, without permission," and that "starting today, our artists and songwriters will share in the revenue generated by Suno's more than 2 million, and growing, paying subscribers." BMG signed last month; signed on Tuesday, one day before launch, and brings TuneCore distribution for tracks made with the licensed models.
Universal and Sony are still suing. Their case, filed in Boston in June 2024, is unresolved; the songwriter Jason Isbell sued separately this month; and a user class action over an unreported data breach is live. The launch also came a day after Suno acknowledged obtaining YouTube audio to train its models while challenging whether Universal and Sony have to complain about the method of acquisition at all. A planned remix feature will be . The payment architecture is the part with the longest tail: labels and publishers receive money from Suno and then decide independently how to allocate it among artists and songwriters, which is the structure that produced two decades of streaming-royalty disputes and leaves an individual songwriter no contractual claim against Suno whatsoever.
Compiled by Matthew Yellin · mattyellin.com · Est. May 2026 · A cross-spectrum brief drawn each morning from wires, filings, and the trade press