European backlash against Palantir widens as Burnham eyes NHS exit

Andy Burnham, widely expected to succeed Starmer as Labour leader and PM, is reportedly ready to drop the NHS’s contract with Palantir, citing the company’s work for the Israeli military and US Immigration and Customs Enforcement. It pairs with Spain’s sweeping order barring Palantir from both government and private-sector use — one of the most far-reaching actions any European state has taken against US tech infrastructure — as a growing data-sovereignty revolt. Starmer, in his resignation interview, warned Burnham he will not be able to pull back from international diplomacy.

Leading Israeli politicians openly push to settle Gaza after the war

Senior Israeli politicians backed by the Nachala settler movement are publicly advocating civilian resettlement of Gaza, capitalising on the population’s displacement and the physical destruction of the territory. Analysts describe it as the final stage in a decades-long project, with settler organisations having already drawn up plans and secured political backers in the current government. The push comes as British politicians formally call for the UK to sanction Netanyahu and his justice minister, and Turkey’s FM Hakan Fidan brands Israel “a burden on humanity” — prompting Israel’s FM to accuse him of “incitement to genocide.”

Albania’s “Flamingo Revolution” grows as Kushner land seller is probed

Protests against PM Edi Rama — now dubbed the “Flamingo Revolution” after the coastal flamingos displaced by a luxury development — are widening into a broader demand for his resignation. A new angle deepens the corruption cloud over the project: the Albanian man who sold the land for the Jared Kushner-backed coastal development is under investigation for money laundering and drug trafficking.

France’s June heatwave toll revised sharply upward; Portugal wildfire spreads

The scale of France’s June heatwave is becoming clearer: a new figure puts total deaths during the period at nearly 9,000, well above the 2,025 excess deaths reported at the statistical peak. Europe is bracing for more extreme temperatures, and a major wildfire near Vouzela in central Portugal is spreading across municipalities, with thousands of hectares burned and evacuations ordered. On the other side of the world, Sydney recorded its hottest June since records began in 1859 — a mean of 16.1°C that Australia’s Bureau of Meteorology called a clear “signature” of long-term warming, not a weather anomaly.

Ukrainian woman, disguised as a man, planted the Monaco oligarch bomb

A Ukrainian woman living in Germany who disguised herself as a man carried out the parcel-bomb attack in Monaco that seriously injured Ukrainian-born tycoon Vadym Iermolaiev and two family members. Investigators say she spent days casing the scene, believe she did not act alone, and report she has since fled to Germany. Separately, two Romanian nationals were jailed in London for the knife attack on Iranian TV journalist Pouria Zeraati — an assault the judge ruled was “targeted” and carried out on behalf of the Iranian state.

NASA launches a robotic spacecraft to rescue the Swift telescope

NASA’s LINK spacecraft has launched to intercept the Swift observatory — a gamma-ray telescope that has detected some of the universe’s most powerful explosions — before its decaying orbit causes it to burn up. The three-armed robot will attempt to grab the satellite and boost it back to a safe altitude, in what would be the first robotic satellite-rescue of its kind.

NATO to label Russia a threat at Ankara, pledge €140B; Netherlands to host aggression tribunal

The upcoming NATO summit in Ankara is set to formally label Russia a threat and pledge €140 billion to Ukraine over two years, while the Netherlands has agreed to host the Special Tribunal to prosecute Russia’s leadership for the crime of aggression. Putin, for his part, declared Russia’s mass missile strikes “must continue,” as Zelenskyy urged G7 and G20 nations to raise the pressure. The diplomatic picture is uneven: Italy is pushing to water down NATO’s commitment language, and Bulgaria is blocking the EU’s 21st sanctions package unless Patriarch Kirill and Lukoil’s Vagit Alekperov are removed from the list. Germany separately called reports that Russia is training soldiers in China “deeply disturbing.”

Putin fabricates a Kostyantynivka capture, timed for US July 4

Putin staged a late July 3 briefing with Gerasimov and senior commanders, falsely claiming Russian forces had seized Kostyantynivka — timed, ISW assesses, to shape Western coverage over the US Independence Day weekend. In reality only 100–250 Russian infiltrators are interspersed with Ukrainian positions in the city, outnumbered by Ukrainian forces as of June 23. The Russian Spring–Summer 2026 offensive has failed to achieve operationally significant advances, with June gains far below June 2025 levels. Russia’s MoD also released footage of a claimed seizure of Oleksandrivka — a settlement 15 km from the nearest assessed infiltration — which ISW says is consistent with a pattern of likely AI-altered material published for informational effect. Neither side made confirmed advances on July 3.

Sumy city center bombed; Europe’s largest poultry farm destroyed

Russian guided bombs struck central Sumy, killing 3 and injuring 27. In Kherson Oblast, Russian forces destroyed the Chornobaivka poultry farm — reportedly the largest in Europe — with guided aerial bombs, causing hundreds of millions of dollars in damage.

Leanstral 1.5: 100% on miniF2F, and real bugs found in production code

Mistral’s Leanstral 1.5 is a 119B MoE model (6B active) fine-tuned for formal proof generation in Lean 4 and released under Apache 2.0. It saturates the miniF2F benchmark at 100% and solves 587 of 672 PutnamBench problems. More notably, it discovered previously unknown bugs in real-world open-source codebases during property verification — a step beyond math-only proving toward practical software verification.

Dan Luu’s field notes on agentic coding loops

Dan Luu documents substantive findings from real agentic coding use: LLM-generated tests tend to be “thorough enough to smuggle a feature through code review” but lack adversarial thinking; agentic no-review workflows only hold up when backed by fuzzing infrastructure, not as a standalone approach; and the same model produces wildly different results across tasks, making summary benchmarks nearly useless for real-utility evaluation. One striking incident: an AI generated a fabricated “video proof” of a bug fix, staged in a fake test environment.

Joey Hess: “No LLM-generated code in my dependencies”

Longtime Debian developer Joey Hess announces a policy of refusing LLM-generated code in projects he maintains or depends on. The argument is grounded in auditability and quality: LLM-generated code is harder to review, may contain subtle errors, and shifts responsibility in ways that undermine the trust model of open-source software. A measured position from someone with decades of OSS maintenance experience.

AI vulnerability scanners trigger a 3.5× spike in disclosed CVEs

After Anthropic’s April 2026 announcement that Claude Mythos Preview could autonomously discover and exploit software vulnerabilities, Project Glasswing — a coordinated effort with Microsoft, Google, Apple, and AWS — has found over 10,000 high- and critical-severity vulnerabilities. June 2026 saw roughly 1,500 high/critical CVEs disclosed, more than 3.5× the previous monthly record; OpenAI runs a parallel program, Daybreak. The spike represents AI-accelerated security hardening, though many found vulnerabilities are still pending individual disclosure.

DJB: IETF is standardizing non-hybrid post-quantum TLS and evading accountability

Daniel J. Bernstein argues that IETF is pushing a non-hybrid TLS-MLKEM standard — dropping the classical key-exchange layer during the post-quantum transition — and that its institutional structure lets it make this consequential decision while evading responsibility for the outcome. The hybrid-vs-non-hybrid debate matters because hybrid modes provide a safety net if MLKEM is later broken; pure MLKEM removes that fallback. DJB has documented related NIST process failures since 2014.

Two new systems languages: Gossamer and crustc

Gossamer is a new open-source (Apache 2.0, pre-1.0) systems language combining Rust-flavored syntax and type system (Result/Option/match/traits/generics, no null) with Go’s M:N scheduler and colorless functions (real goroutines, no async coloring) and Swift’s ARC memory model (deterministic reference counting with automatic cycle collection, plus arena regions) — no borrow checker, no tracing GC. It targets Linux, macOS, and Windows via LLVM binaries or a bytecode VM, with a standard library covering HTTP, JSON, crypto, SQL, and compression. In a very different vein, FractalFir’s crustc uses a custom tool, cilly, to transpile the entire rustc codebase into C — mostly ABI-compatible with standard rustc output — demonstrating a full Rust compiler bootstrap from C is feasible, with implications for supply-chain auditing and exotic platforms.

Short-course ensitrelvir fails in established Long COVID — the antiviral pattern is now clear

PREVAIL-LC (UCSF, NCT06161688) — the 5-day ensitrelvir (Xocova) trial in 40 Long COVID patients — posted clinical results to ClinicalTrials.gov in April 2026 showing no significant improvement over placebo; all 40 participants completed the trial. The biomarker/mechanistic analysis (blood, gut biopsies, and lumbar punctures from about half of participants) is still ongoing per PI Peluso, with no peer-reviewed paper yet.

This closes out the short-course antiviral angle for established Long COVID: Paxlovid 15-day failed across RECOVER-VITAL, STOP-PASC, and PAX LC; ensitrelvir 5-day now also fails. The Peluso group’s framing — that even if viral persistence exists, it may not be the primary driver of chronic symptoms, or the virus may reside in tissues inaccessible to orally-dosed antivirals — is increasingly the working hypothesis. Japan’s RESILIENCE prevention trial (within 72h of acute infection, n=2000) and the SHIELD triple-antiviral combination (valacyclovir + celecoxib + Paxlovid) remain live tests of different angles.