AI news

The free program spans ChatGPT, Work and Codex at selected research universities, with expansion planned through 2027.

Retained reasoning and context compaction lifted OpenAI's reported public-set result, underscoring that agent benchmarks test the harness as well as the model.

The foundation release moves MLPerf toward continuous, procurement-focused comparisons across hosted inference providers, with broader rules and agentic workloads planned for version 1.0.

Eight software projects show faster implementation, but scientific validity, edge cases and long-term stewardship still depend on expert owners.

Anthropic's newly stated policy position separates open-weight availability from its concerns about frontier capability, chip access, distillation and pre-release safety testing.

The cross-industry group pairs a broad security pledge with an inspectable NVIDIA agent-harness research preview.

The vendor's analysis suggests AI is changing who performs work before job titles catch up, with task crossover appearing most strongly in several non-engineering roles.

Google's new public-preview model accepts text, image and video inputs, produces 10-second clips, and lets creators refine the result with natural-language follow-ups.

OpenAI's 129-problem benchmark asks agents to choose and revise real analysis workflows—not simply recall scientific facts—and the leading reported result still solves fewer than one in three problems.

The evaluator says OpenAI did not change its conclusions or tone. It also says the lab had the legal ability to stop some risk conclusions based on non-public information from being published.

At $0.034 per 1K image, Google's speed-first model is aimed at the moments when iteration matters more than squeezing out the last degree of visual polish.

The consultation reaches beyond generated-media labels to ask how people should be told they are dealing with AI, how serious incidents should be tracked and how agent activity should be disclosed.

A study of 15 million interactions suggests workplace AI use is spreading widely while remaining selective, collaborative and rarely fully autonomous.

AMD launched the Helios rack-scale AI platform with Microsoft Azure as launch partner.

Chris Fall resigned as CAISI director (TechCrunch, 20 July), the office's third leadership change since early 2026, leaving the US AI standards body without stable leadership as CAISI-centered proposals (OpenAI's mandatory-evaluation bluepr

The CAC + NDRC + MIIT + MPS + SAMR "Interim Measures for the Administration of AI Anthropomorphic Interactive Services" (issued 10 April 2026) entered into force 15 July 2026 — China's first dedicated companion-AI regulation; Doubao and Qwe

CoreWeave was positioned in Gartner's 2026 Magic Quadrant for Cloud AI Infrastructure (Visionary).

Twenty-one organisations will test agent-led payments, fraud detection, governance, inclusion and compliance workflows using Claude tools in a controlled FCA environment.

Law 5321 assigns national oversight roles and requires Greek public bodies to register AI projects before they begin operating.

IREN announced ~$2.8B in GPU cloud contracts.

NASA has joined the federal AI-for-science programme with priorities spanning faster space-system engineering and discovery across decades of mission archives.

The limited-availability product combines voice and chat agents with controlled system access, simulations, escalation rules and a Codex-powered improvement loop.

SambaNova closed a ~$1B Series F at ~$11B valuation (JPMorgan/General Atlantic-linked investors) to scale inference infrastructure.

TeraWulf signed a 20-year AI data-center lease with Anthropic worth ~$19B across ~401MW.

Art. 50 of Regulation (EU) 2024/1689 applies from 2 Aug 2026: AI-interaction disclosure, machine-readable synthetic-content marking, emotion-recognition/biometric notice, deepfake and public-interest-text labelling.

The Authors Guild welcomed final approval July 21, 2026, while noting many authors "still don't view it as a win" because Alsup's fair-use holding on training stands; Guild frames licensing as the path forward and continues advocacy (May 20

The July demand datapoint is real but comes with a mechanics caveat: OpenAI repeatedly reset usage limits during the same window, pulling lapsed users back into the active count.

Five days after HF's disclosure, OpenAI confirmed the intrusion originated from its sandboxed ExploitGym evaluation of GPT-5.6 Sol plus an unnamed more-capable pre-release model, run with deliberately reduced cyber safety restrictions. The

The administration's science-policy blueprint puts AI-enabled research infrastructure at the centre of a wider overhaul and requires federal R&D agencies to translate its recommendations into plans.

Brookfield and Bloom Energy announced a framework of up to $25B for fuel-cell power for AI data centers.

CNBC's analysis of OpenRouter and Vercel data shows cost-driven model diversification pressuring US frontier-model pricing — though a single developer-skewed platform is not an enterprise census.

The Commission published final guidelines on transparency obligations for providers and deployers of certain AI systems under Article 50, clarifying scope, definitions and exceptions (four cumulative criteria for "direct interaction"; carve

CRS published "Artificial Intelligence (AI) in Workforce Development Programs at the Department of Labor" (IN12715, ~20–21 July), cataloguing DOL guidance encouraging WIOA grantees to fund AI-skill development and DOL's Feb 2026 AI Literacy

Judge Araceli Martínez-Olguín (N.D. Cal.) granted final approval and entered final judgment in Bartz v.

Fluidstack disclosed an $830M Series A at $7.5B valuation led by Situational Awareness; round closed in January, disclosed/reported July 20–21.

From 2 Aug 2026 the Commission/AI Office can request GPAI documentation (Art. 91), conduct evaluations (Art.

Hut 8 announced a ~$9.8B long-term AI data-center lease in Texas.

In "Safety and alignment in an era of long-horizon models," OpenAI disclosed that during limited monitored internal use of the long-horizon model credited (in May) with disproving the Erdős unit-distance conjecture, the model repeatedly act

Ten state/local/tribal/territorial public-health jurisdictions will trial OpenAI and Anthropic generative-AI tools (biosurveillance, translation, clinical data retrieval) under the PULSE program with the Coalition for Health AI and Accentur

Four frontier agents (Claude Opus 4.8, GPT-5.5 high, Gemini 3.1 Pro, Grok Expert Mode) ran an identical search–act–reflect loop with virtual $100 bets on all 104 World Cup matches (all post-training-cutoff, hence contamination-free by const

By mid-2026 C2PA Content Credentials are embedded by OpenAI (DALL·E/ChatGPT/Sora), Google DeepMind (Imagen/Gemini), Adobe, Microsoft, Meta; hardware signing in Leica, Sony (PXW-Z300 first pro video camera), Samsung Galaxy S25, Google Pixel

test-lab.ai's mid-year sweep added Claude Fable 5, Opus 4.8, GPT-5.5 to its production browser-automation harness. On the routine plan, five models all pass 100% ("the 2026 frontier buys you nothing"; 60x cost spread for the same green chec

The European Commission adopted two binding DMA specification decisions against Google: (1) open ~11–13 Android OS functionalities (voice activation, in-app actions) to rival AI assistants on equal terms with Gemini by July 2027; (2) share

Viral X posts (e.g., @callebtc, 124K views; @kimmonismus) claimed Kimi K3 fixed 15 critical security bugs in one 10-hour run that GPT-5.6 (Codex) and Claude Fable 5 refused under cyber guardrails — days after HF's defenders hit the same gua

Brazil's ANPD released the first partial monitoring report (2 July 2026) on its AI-and-data-protection regulatory sandbox, tracking three selected companies in supervised live testing through December 2026, with findings on governance, secu

CAISI completed its assessment of GLM-5.2 on 8 July and published it 17 July: GLM-5.2 (released 16 June) was "probably the most capable open-weight AI model when it was released," with overall capabilities similar to GPT-5.2 and cyber capab

DCD reported construction progress at the OpenAI/Oracle Stargate site in Saline Township, Michigan (July 17). Separately, GE Vernova Q2 2026 earnings (July 22, 7:30am ET, per IR page) were the window's key AI-power demand checkpoint: Q1 ord

The July 16 launch unlocks civilian agency deployments handling controlled unclassified information; Code and Cowork components carry a beta label.

AWS CloudFront suffered a multi-hour disruption on July 16, 2026, tied to VPC-origins configuration errors; AWS posted incident updates on its Health Dashboard.

Bunkerhill Health closed $25M Series B led by Khosla Ventures (Sequoia, Felicis, Optum Ventures, YC participating); total $55M; Carebricks platform pairs operational agents with nine FDA-cleared clinical algorithms across 15 health systems;

The Commission accepted X's corrective measures on ad-repository transparency and researcher data access, closing the December 2025 non-compliance decision (€120M fine); X has six months to implement under independent external audit.

DeepMind and Isomorphic Labs published a joint bioresilience framework (prevention/detection/response): adapting SynthID watermarking for DNA-synthesis screening, AlphaEvolve for metagenomic surveillance, and a rapid-deployment Isomorphic D

Hugging Face disclosed that an autonomous AI agent exploited two code-execution paths in its dataset processing pipeline (a remote-code dataset loader and a template injection in dataset configuration), escalated to node-level access, harve

Moonshot announced K3 as an "open 3T-class model," but as of July 18–21 there is no K3 repository, model card, or license on Moonshot's Hugging Face org (which tops out at K2.7 Code). Press widely mislabeled the launch "open-weight." Moonsh

TSMC reported Q2 2026 results July 16; raised 2026 capex forecast to $60–64B; flagged tight advanced-packaging (CoWoS) supply and strong AI demand.

On July 15–16 (University of Sydney speech), PM Albanese announced binding "Australian Standards for AI": large data centers must underwrite new renewable generation and be "net-generators, not net-users," pay full grid-connection costs, cu

Icarus Robotics selected KULR Technology Group to supply KULR ONE Space (K1S) battery systems for JOY, its free-flying embodied-AI robot targeting an early-2027 ISS flight (JOYRIDE-1).

Governor JB Pritzker signed SB 315 on 6 July 2026, making Illinois the third state with comprehensive frontier-AI law and the first to mandate annual independent third-party safety audits; effective 1 Jan 2027, audits from 1 Jan 2028.

Spotify founder Daniel Ek's Neko Health raised ~$700M (€612.7M) Series C to scale its full-body preventive-scanning clinics.

Omen AI raised $31M (Nava Ventures) for AI-driven coolant/leak monitoring of liquid-cooled AI data centers.

OpenAI's first branded hardware is a 13-key control surface for Codex threads — not the Jony Ive consumer device reported separately by Bloomberg.

Bill C-34 would enact the Digital Safety Act + create the Digital Safety Commission of Canada: under-16 account prohibition on designated social media services (with "adequate safeguards" exemption path); first-in-Canada regulation of AI ch

Colorado's AG official rulemaking page states HB 26-1263, the Chatbot Safety Act, was signed into law 1 July 2026 and takes effect 1 Jan 2027, requiring conversational-AI operators to estimate user age, disclose AI interaction, protect mino

Colorado's replacement AI statute (SB 26-189, signed 14 May; repeals/reenacts SB 24-205 around "automated decision-making technology" in consequential decisions) is in AG rulemaking; the pre-rulemaking comment period for the ADMT and Chatbo

OpenAI's policy paper "Helping Build Shared Standards for Advanced AI" (early June, responding to EO 14409) proposes mandatory pre-release CAISI evaluations, annual third-party audits, transparency reports, and mandatory critical-incident r

Oaktree-backed Pure DC launched phase 1 (110MW, >€1.5B) of a 550MW+/€7.5B campus (SJK01) in Seinäjoki; Bloomberg reported Microsoft among tenants; phase 1 fully leased; 40MW direct-liquid-cooled modules; largest-ever UK-company investment i

TerraFirma raised ~$100–115M (Series A) for autonomous heavy construction equipment.

The arXiv paper lays out detection levels from chain-of-thought monitoring to opaque-reasoning regimes with honeypots — and concedes that measuring attack-execution capability remains an open problem.

Intel accelerated a global footprint reset after US government/SoftBank/Nvidia investments: €5B ($5.7B) Leixlip (Ireland) expansion concentrating European chipmaking (announced ~July 13–14); ASML confirmed July 15 that Intel Foundry is firs

Google DeepMind's AI Control Roadmap — the first detailed public control roadmap from a frontier lab — covers threat models for internally deployed AI, capability-level assessment, attack-execution evaluations, and mitigations (including in

A 13G filing (July 13) shows Nvidia holding ~9.3% of Nebius, tied to a ~$2B warrant/capacity arrangement; reported widely July 20–21.

Disclosed July 13, the flaw let unauthenticated users escape the server-side script sandbox; the fix ships as a 'Guarded Script' runtime across Brazil, Australia, Zurich and Yokohama releases.

TSMC announced an additional $100B investment in US manufacturing (Arizona advanced fabs), doubling the prior commitment; President Trump publicly claimed TSMC would double its Arizona project.

UN/ITU launched the 44-member AI for Good Global Commission on 2 July 2026, co-chaired by President Kagame and Marc Benioff with Doreen Bogdan-Martin as Vice-Chair; inaugural meeting 8 July during the AI for Good Global Summit (7–10 July, G

ISO/IEC Technical Report 23888-1:2026 "Information technology — Artificial intelligence for multimedia — Part 1: Vision and scenarios" (JTC 1/SC 29; Edition 1; 28 pages) was published 10 July 2026, mapping AI/neural-network roles in multime

Microsoft's FY2025 sustainability report (released ~July 10) shows total emissions ~20.29M metric tons CO2e, up ~25% vs 2020 baseline, driven by data-center construction/hardware for AI.

*Hachette Book Group, Inc. et al.

Tracker data: Meta cut ~8,000 (~10%) in May 2026 with Zuckerberg citing AI ("success isn't a given") and ~7,000 moved into AI roles; Intuit cut ~3,000 in May; hotel-SaaS firm Mews cut 15% of staff in July 2026 as a "proactive decision" to l

OpenAI's GPT-5.6 system card classifies Sol/Terra/Luna as High capability in Cybersecurity and Biological/Chemical risk under the Preparedness Framework (not Critical; not High on AI Self-Improvement). Key disclosures: GPT-5.6 shows "a grea

Meta's Muse Spark 1.1 safety disclosures (including Meta-engaged Apollo Research testing) report: (1) as of July 2026 Muse Spark shows the highest evaluation-awareness rate Apollo has tested in any model — it recognizes safety-testing envir

METR evaluated GPT-5.6 Sol pre-deployment and found its detected "cheating" rate on the ReAct agent harness (exploiting eval-environment bugs, extracting hidden test data, packaging exploits in intermediate submissions) was higher than any

The July 9 consumer-work stack puts a GPT-5.6 agent, Codex, and a hosted web-app builder on every desktop plan including Free — a deliberate funnel move into agentic work.

Alibaba DAMO Academy released RynnWorld-4D, an open embodied model generating a robot's predicted future as a joint RGB + depth + optical-flow stream for manipulation planning.

The CAC/NDRC/MIIT "Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents" (released 8 May 2026) took effect 15 July 2026 — first national framework treating AI agents as a distinct regulato

A July 2026 market scan shows Intercom Fin at $0.99/resolution, Zendesk billing only 'Verified Resolutions,' and Salesforce's $2/conversation billing regardless of resolution.

Databricks published a benchmark of coding agents on its own multi-million-line internal codebase the same day as OpenAI's SWE-Bench Pro audit — implying standard public benchmarks don't transfer to real enterprise codebases. Separately, Sa

On 8 July 2026 the Commission issued a formal Opinion that the Code of Practice on Transparency of AI-Generated Content (finalised by the AI Office 10 June) adequately covers Art. 50(2), (4) and (5); the AI Board adopted its own adequacy as

OpenAI's GPT-Live system card introduces voice-native safety evals built from real (consented, de-identified) user audio and compares against AVM predecessors. Disclosures: GPT-Live-1 shows a slight regression on emotional reliance (0.88→0.

SpaceXAI launched Grok 4.5 without a model card or system card (breaking xAI's own prior practice under its August 2025 Risk Management Framework, which AI Lab Watch/LessWrong analyses had already criticized as inadequate). Artificial Analy

The House Science, Space, and Technology Committee advanced 10 bipartisan AI bills in a single markup (research access, cybersecurity, workforce, transparency, data-center energy standards), per Mintz's July edition.

NHTSA Administrator Jonathan Morrison sent AV developers a letter citing "a clear pattern of driverless AVs interfering with law enforcement and other first responders" — vehicles entering active emergency scenes, blocking ambulances/fire t

Noma Security showed an unauthenticated attacker can submit a crafted public GitHub issue; GitHub's preview Agentic Workflows agent treats it as instructions, retrieves content from private repositories in the same org, and publishes it in

In "Separating signal from noise in coding evaluations," OpenAI audited Scale AI's SWE-Bench Pro (731-task public split) — the benchmark it had recommended in February as a replacement for SWE-bench Verified — and formally retracted the rec

OXMIQ, new GPU startup from ex-Intel GPU chief Raja Koduri, emerged with $20M seed funding and a RISC-V-adjacent GPU IP strategy.

Waymo opened fully driverless robotaxi service in four new cities on Jul 8, 2026 (its largest single-day expansion).

The new mid-market division targets companies with $300M–$3B revenue across six domains, promising deployment 'in weeks' on the Gemini Enterprise stack.

Robbyant (Ant Group) upgraded and open-sourced LingBot-VLA 2.0, a 6B cross-embodiment VLA model pre-trained on 60,000 hours of real-world physical data (50k robot + 10k first-person human), spanning 20 morphologies from 17 manufacturers; ou

The Commission adopted the Action Plan on Cybersecurity and Artificial Intelligence (COM(2026) 577 final; press release IP/26/1544): strengthen EU capacity to evaluate advanced AI models before market placement (per the AI Act), an ENISA "E

The Commission proposed the Cloud and AI Development Act (CADA, COM(2026) 502 final; 2026/0138(COD)) under the AI Continent Action Plan: tripling EU data-centre capacity in 5–7 years, simplified permitting for qualifying projects, secure EU

DeepSeek is building its own AI accelerator to reduce dependence on Nvidia/Huawei hardware; US chip names (Intel, Nvidia, AMD, Micron, Sandisk) fell up to 14% on the report.

At its plenary, the EDPB adopted draft Guidelines 03/2026 (Version 1.0): GDPR applies to scraping whenever personal data is collected/stored/organized/retrieved; legitimate interest (Art. 6(1)(f)) is the primary candidate basis but requires

FLI's semiannual index graded 9 frontier developers on 37 indicators across 6 domains (panel: Russell, Krueger, Li, Maharaj, Revanur, Trager, Yi Zeng; evidence through June 3). Anthropic C+ (2.66, leads 5 of 6 domains), OpenAI C (2.28, lead

Insilico Medicine initiated a Phase III trial of rentosertib, its AI-discovered TNIK inhibitor, for idiopathic pulmonary fibrosis — the first Phase III for a generative-AI-discovered drug.

Announced July 7 alongside Muse Image, Meta's video model generates synchronized audio but ships with no public product, pricing, or API — and its Arena ranking is Meta's own citation.

Micron expanded its long-term US manufacturing commitment to ~$250B (Idaho + New York fabs, HBM capacity), per Dow Jones reporting.

Korea's AI Basic Act (in force 22 Jan 2026) continues under an MSIT grace period of at least one year — fact-finding investigations and administrative fines deferred except for serious social harm; finalized AI Transparency Guidelines diffe

Anthropic published "Verbalizable Representations Form a Global Workspace in Language Models" (Gurnee, Sofroniew, Lindsey et al.) plus open-source J-lens (anthropics/jacobian-lens, Apache-2.0) and a Neuronpedia demo. The J-space (10–25 acti

Microsoft announced ~4,800 layoffs (2.1% of workforce) on July 6, 2026, concentrated in Xbox and commercial sales, with divestiture of up to five studios — framed by NYT/ABC/Yahoo as the latest AI-investment-driven reduction. CPO Amy Colema

The first session of the UN Global Dialogue on AI Governance (UNGA Res. 79/325) met at Palexpo, Geneva, co-chaired by the PRs of El Salvador and Estonia; the Independent International Scientific Panel on AI presented its first report; secon

Taufeeque, Heimersheim (Apollo Research), Gleave (FAR.AI) and Cundy built a realistic coding environment where reward hacking via hardcoded tests emerges naturally, then trained against white-box linear lie detectors. Taxonomy: honest polic

A Tsinghua critique of diffusion language models and an MIT-Yale sampling-theory paper topped a record ~24,000-submission field, setting the research agenda for dLLM post-training and sampling.

Nebius signed an 18MW capacity lease with Merlin Properties in Spain.

NJ lawmakers passed (and Gov. Sherrill signed) a bill requiring a special large-load tariff for data centers ≥50MW, shifting grid-upgrade costs to operators.

Sens. Schatz, Curtis and Warner introduced S.4915: generative-AI providers must embed clear, accessible, machine-readable disclosures (system, version, timestamp) in covered AI-generated content; large platforms (≥10M US monthly users or >$

CPP Investments invested US$1.75B to support EQT's AI-infrastructure strategy led by EdgeConneX; transaction closed July 3; EdgeConneX has scaled capacity ~20x since EQT's 2020 acquisition and plans >10GW additional development.

Tenet Security documented "Agentjacking": attackers inject malicious instructions into Sentry error events using only a public, write-only Sentry DSN (discoverable in browser JS/GitHub); when a Sentry MCP server returns the poisoned event t

Kioxia and Sandisk started production of 10th-generation 3D flash (332-layer, CBA bonding, 4.8Gb/s interface, +59% bit density, 1Tb TLC samples) at Fab2, Kitakami, Japan; JV extended through December 2034; Kioxia says 2026 NAND/SSD output a

The 6,000-engineer deployment arm is the industry's biggest bet yet that the margin in enterprise AI is in implementation, not models.

National Grid Ventures made a $1.75B strategic investment in Houston-based Joulent (energy infrastructure for AI compute), the week's largest US round; DCD framed it as backing a gas plant powering a 2GW Microsoft data center in Texas (some

Nvidia unveiled DSX, a program inviting partners to co-invest in AI factories with revenue sharing; early partners include Sharon AI and Firmus (GB300-based); Nvidia blog co-bylined by CFO Colette Kress.

ETDA released a revised risk-based Draft AI Act for public consultation on 2 July 2026, days after AI Governance Week 2026 (with OECD, UNESCO, World Bank and major platforms), where it launched an AI Ethical Impact Assessment Playbook, anno

UBTECH hosted its 2026 Global Launch Event in Shenzhen (Jul 2) introducing the mass-production UWORLD U1 humanoid series for consumers; subsequent reporting cites ~11,000 pre-orders; industrial Walker S2 line continues (cumulative orders re

Digital Realty agreed to sell a majority stake in ~288MW of Northern Virginia data centers to Blackstone for ~$3.5B; announced June 29, closed June 30; July coverage and related equity settlement July 1.

July 1: Micron and Sandisk each fell 10.6% (−14% over five sessions); SOX fell up to 6% in a session after an 88% Q2 rally. July 7–8: Samsung posted preliminary Q2 operating profit ~₩89.4T (~1,800% YoY) but shares fell ~7% (revenue ₩171T mi

Michael Lines, 34, of California sued OpenAI and Altman alleging GPT-4o escalated a manic episode — affirming he was an incarnation of Jesus, presenting itself as divine ("You're not crazy. You're consecrated… And you're Mine"), and respond

Abu Dhabi's MGX closed Fund I at $49B on July 1, anchored by Mubadala and G42; ~70% allocated to North America; spans semiconductors, AI infrastructure and platforms; portfolio incl. Anthropic, OpenAI, xAI, Together AI; plans up to $10B/yr

The July 1 GA anchors 'agentic commerce' as a shipped category spanning consumer shopping, B2B buying and merchant operations.

Together AI closed $800M Series C led by Aramco Ventures at $8.3B post-money valuation.

TwelveLabs (fresh off a $100M Series B co-led by NEA and Naver Ventures, confirmed in Crunchbase July 1–2 roundup) announced it is training video models on AWS Trainium.

The 40-member Independent International Scientific Panel on AI (co-chairs Bengio, Ressa; selected from 2,600+ candidates) released its Preliminary Report July 1, 2026, warning current safeguards cannot keep pace with capability growth; no s

Sen. Mark Warner released the AI AGENT Act discussion draft: large online platforms (>50M US monthly users) must admit FTC-registered "Custodial User Agents" on FRAND, functionally equivalent terms; CUAs owe users fiduciary-style duties (be