AI News Feed

CNBC confirmed on June 24, 2026 that Anthropic sent a June 10 letter to the US Senate Banking Committee—addressed to Sens. Tim Scott and Elizabeth Warren—accusing Alibaba of attempting to 'brazenly' and 'illicitly' extract its AI capabilities. Anthropic says operators affiliated with Alibaba and its AI lab ran 28.8 million exchanges with Claude models using roughly 25,000 fraudulent accounts, calling it 'the largest known distillation attack on Anthropic to date.' Distillation trains a smaller model on a stronger one's outputs; the disclosure follows Anthropic's February claims of 'industrial-scale' campaigns tied to DeepSeek, Moonshot, and MiniMax.

Added: June 23, 2026

CNN's June 21, 2026 analysis lays out how Anthropic became the test case for an improvised US AI export-control regime. After the Department of Defense labeled Anthropic a 'supply-chain risk' over a guardrails dispute, the Commerce Department issued a June 12 order barring distribution of Fable 5 and Mythos 5 to foreign nationals worldwide—forcing both models offline globally on roughly 90 minutes' notice. Civil-liberties and cybersecurity groups warn the precedent treats frontier models like weapons systems requiring government clearance to ship.

Added: June 20, 2026

Bloomberg reporting amplified through mid-June 2026 details how the US Defense Department stood up an evaluation pipeline—separate from the Claude-powered Maven Smart System—to test models from OpenAI, Google, and xAI as replacements for Claude across classified networks. The rupture followed Defense Secretary Pete Hegseth's early-March designation of Anthropic as a 'supply-chain risk' after the company refused to drop guardrails barring mass surveillance and fully autonomous weapons. Insiders peg a realistic replacement timeline at 12+ months given Claude's deep AWS integration.

Added: June 16, 2026

NPR reported on June 22, 2026 that the AI industry has poured money into a Manhattan Democratic primary, turning the NY-12 race to succeed retiring Rep. Jerry Nadler into a proxy war over future AI regulation. State Assemblymember Alex Bores—author of New York's RAISE Act—has been targeted by the OpenAI-aligned super PAC network Leading the Future, which has spent $7.6M+ against him from a $100M+ war chest backed by Andreessen Horowitz, Greg Brockman, Joe Lonsdale, and Perplexity. Pro-regulation Anthropic is helping bankroll the other side. Groups tied to the two labs have together spent $15M+ on the race, decided in the June 23 primary.

Added: June 21, 2026

Sensor Tower's State of AI 2026 report (released June 16, 2026) finds ChatGPT's share of global AI-assistant users fell to 46.4% by the end of May—the first time below 50% since the category began in November 2022. Google's Gemini climbed to 27.7% and Anthropic's Claude reached 10.3%. ChatGPT now has 1.1B+ monthly active users, Gemini 662M, and Claude 245M. Sensor Tower attributes Gemini's rise to OS-level Android integration and notes Claude leads monetization, with 13% of users paying—the highest conversion rate among major assistants.

Added: June 15, 2026

On June 16, 2026, SpaceX announced an all-stock deal to acquire Anysphere—the startup behind the AI coding agent Cursor—for $60 billion, days after a blockbuster Nasdaq debut. Billed as the largest acquisition of a VC-backed startup in history, the deal slots Cursor into the xAI ecosystem (SpaceX absorbed xAI in a February merger), letting Cursor shift from reliance on Anthropic and OpenAI APIs toward xAI's Colossus compute and in-house Grok/Composer models. Cursor reached ~$4B annualized revenue and 50,000+ enterprise clients; the deal is expected to close in Q3 2026 pending regulatory approval.

Added: June 15, 2026

On June 18, 2026, Noam Shazeer—VP of Engineering and co-lead of Gemini, and a co-author of the 2017 'Attention Is All You Need' paper that introduced the Transformer—announced he is leaving Google to join OpenAI as Lead for AI Architecture Research, overseeing next-generation model design and the evolution of (or successors to) the Transformer. The move is a notable talent win for OpenAI as it prepares a confidential IPO filing, and a setback for Google less than two years after its ~$2.7B Character.AI deal brought Shazeer back. Sam Altman called him 'one of the people I have most wanted to work with since the very beginning of OpenAI.'

Added: June 17, 2026

CNBC reported on June 17, 2026 that Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis used a closed-door meeting at the G7 summit to call for a US-led AI coalition. Around a dozen tech executives—including OpenAI's Sam Altman—joined heads of state to discuss the opportunities and risks of increasingly capable AI. The meeting followed the release of frontier models with cyber capabilities advanced enough that some experts warn of major disasters if misused, and surfaced a possible 'trusted-partners' framework to let approved countries access advanced US models despite recent export restrictions.

Added: June 16, 2026

Bloomberg reported (June 9, 2026, amplified through late June) that China is preparing to spend roughly 2 trillion yuan ($295 billion) over five years to build a unified national network of interconnected data centers, targeting full interconnection by 2028. The most consequential feature is a domestic-sourcing mandate requiring at least 80% local technology—favoring Huawei, Biren, Moore Threads, and Alibaba Cloud chips and effectively squeezing out Nvidia and AMD. The figure excludes private spending by Alibaba and Tencent; including grid integration, total investment could reach ~$740B.

Added: June 11, 2026

A June 2, 2026 White House executive order established a framework where developers can voluntarily engage the federal government to determine whether a model meets the 'covered frontier model' designation and provide pre-release access for up to 30 days to trusted partners. The order also calls for an AI cybersecurity clearinghouse—in voluntary collaboration with industry and critical-infrastructure operators—to coordinate vulnerability scanning, validation, and patch distribution. Importantly, the order does not authorize any mandatory licensing, preclearance, or permitting requirement for developing or releasing AI models.

Added: June 1, 2026

In a unanimous June 18, 2026 vote, the Federal Energy Regulatory Commission issued tailored Section 206 'show cause' orders to all six grid operators under its jurisdiction—PJM, MISO, SPP, CAISO, ISO New England and NYISO—directing them to justify or overhaul how data centers and other large loads connect to the grid. Operators have 60 days to defend or reform tariffs across five areas and 30 days to report how they will ensure adequate generation. Chair Laura Swett called the urgency a 'national priority'; the proceeding touches markets serving ~200 million Americans.

Added: June 17, 2026

On June 22, 2026, OpenAI moved Aardvark—its GPT-5-based agentic security researcher—into Codex as 'Codex Security' in research preview, rolling out to ChatGPT Enterprise, Business, and Edu via Codex web with the first month free. After scanning the open-source repos it depends on, OpenAI says maintainers' real problem isn't too few vulnerability reports but too many low-quality ones, so the system is tuned to surface high-confidence, actionable issues and cut false positives. OpenAI plans free coverage for select non-commercial open-source projects under a coordinated-disclosure framework; Aardvark has already earned 10 CVEs on open-source software.

Added: June 21, 2026

On June 8, 2026, Microsoft AI announced seven new models developed in-house, spanning image, voice, transcription, coding, and reasoning to form the multimodal MAI model family—including MAI-Code-1-Flash, its first dedicated coding model, available in GitHub Copilot and VS Code. CEO Mustafa Suleyman said that after tuning for consulting firm McKinsey, the coding model outperformed OpenAI's GPT-5.5 with roughly 10x better cost efficiency. The same announcement highlighted a Mayo Clinic healthcare collaboration, underscoring Microsoft's strategy to reduce dependence on partner models and compete across the frontier stack.

Added: June 7, 2026

The New Stack reports that after a year of competition, the leading agentic coding tools have settled on a shared blueprint—orchestration, execution, and review layers—rather than one tool winning outright. Cursor 3.7 ships Composer 2.5 with parallel cloud agents; OpenAI published an official plugin that runs inside Anthropic's Claude Code; and early adopters now run all three together. Because agents are increasingly interoperable through MCP and ACP, the model layer underneath is where most of the cost and output quality actually lives—every agent is a harness around a model.

Added: June 19, 2026

Morph's mid-June 2026 agent leaderboard shows Codex CLI with GPT-5.5 leading Terminal-Bench 2.1 at 83.4%, with Claude Code on Opus 4.8 close behind at 78.9% (Fable 5 scored higher before going offline). The state of the art has moved fast—Claude Opus 4.8 now fixes 69.2% of SWE-bench Pro GitHub issues, versus sub-40% just two years earlier. The leaderboard's authors caution that agents still fail on long, underspecified, or unfamiliar tasks and work best with human review on each change rather than fully unattended, and that benchmark scores should be validated against your own workload.

Added: June 17, 2026

Tenet Security's Threat Labs disclosed 'agentjacking,' an attack (publicized mid-June 2026) in which an attacker uses only a public, write-only Sentry DSN—often scraped from frontend JavaScript or GitHub—to inject markdown-formatted instructions into an error event. When a developer asks an AI coding agent like Claude Code, Cursor, or Codex to 'fix unresolved Sentry issues,' the agent retrieves the poisoned event via the Sentry MCP server and executes the embedded commands with the developer's own privileges, exfiltrating AWS, npm, Docker, and git credentials. Researchers found 2,388+ injectable organizations and an ~85% success rate; Sentry added a content filter but called the root cause a model-side trust problem.

Added: June 15, 2026

OpenAI announced on June 21, 2026 that Samsung Electronics will deploy ChatGPT Enterprise and Codex to its ~125,000 employees in South Korea and to every worker in its global Device eXperience division—one of OpenAI's largest enterprise rollouts. Codex won't be limited to engineers: marketing, product, manufacturing, and corporate-support staff will use it for automation and app-building. The move reverses Samsung's 2023 blanket ban on generative AI (imposed after a source-code leak), following an April–May proof-of-concept and a new control granting access only after internal security training. Samsung SDS also becomes the first Korean entity authorized to resell ChatGPT Enterprise.

Added: June 20, 2026

VentureBeat reports that Z.ai (formerly Zhipu) released GLM-5.2—a ~753B-parameter sparse-MoE model (~40B active per token) with a 1M-token context—under a permissive MIT license, with weights on Hugging Face and ModelScope. Independent Artificial Analysis scoring places it atop open models, and on SWE-bench Pro it scored 62.1, beating GPT-5.5 (58.6). The June 13 launch was timed two days after the Fable 5 export ban, with Z.ai emphasizing 'no regional limits.' Caveat: the hosted API routes prompts through Z.ai servers subject to Chinese law—self-hosting (≈1.5TB of GPU memory) avoids that exposure.

Added: June 12, 2026

The June 2026 open-weight wave is the most compressed in the category's history. DeepSeek V4-Pro posts 80.6 on SWE-bench Verified, 93.5 on LiveCodeBench, and a 3,206 Codeforces rating—leading several benchmarks including closed frontier APIs—and is notable for training at 1.6T-parameter scale on Huawei Ascend rather than Nvidia. Moonshot AI's Kimi K2.7 Code (June 12, 1T params/32B active, 256K context) and MiniMax M3 (June 1, 428B/23B active, ~15x faster decode at full context) round out the field. Independent SWE-bench numbers still lag vendor claims, so treat figures as promising rather than proven.

Added: June 12, 2026

OpenAI published (June 23, 2026) how Derya Unutmaz, an immunologist at The Jackson Laboratory for Genomic Medicine, used GPT-5 Pro to revisit a three-year-old puzzle: how glucose shapes the way T cells develop and specialize. The model surfaced novel, literature-consistent hypotheses across age groups—and in a striking test, correctly predicted the outcome of an unpublished experiment showing CD8+ T cells' enhanced ability to kill lymphoma cells, results it could not have found online. Unutmaz now describes such models as research 'collaborators,' a case study in AI-augmented, expert-validated scientific discovery rather than AI replacing scientists.

Added: June 22, 2026

Salesforce announced on June 15, 2026 a definitive agreement to acquire Fin—the customer-agent company formerly known as Intercom—for approximately $3.6 billion, folding its technology and team into the Agentforce portfolio. Fin's AI support agent handles conversations end-to-end across chat, email, phone, WhatsApp, SMS, and Slack, is powered by a purpose-built model the company calls Apex, and reportedly resolves about 76% of incoming support requests without a human, bringing 30,000+ business customers to Salesforce. The deal is expected to close in Q4 of Salesforce's FY2027.

Added: June 14, 2026

AWS announced (mid-June 2026, with the Bedrock listing updated June 24) that xAI's Grok 4.3 is generally available on Amazon Bedrock, marking xAI's formal entry as a Bedrock model provider and letting enterprises use the model without a separate xAI account. Grok 4.3 offers a 1-million-token context window (30K max output), configurable reasoning effort, tool calling, structured output, and streaming, and claims the lowest hallucination rate among frontier models. On-demand pricing is $1.25 per million input tokens and $2.50 per million output ($0.20 cached input); Grok is now also available on Azure AI Foundry, Oracle Cloud, and Databricks.

Added: June 23, 2026

The Reuters Institute's Digital News Report 2026 (published June 18, based on ~100,000 interviews across 48 markets) finds weekly use of AI chatbots for news rose from 7% to 10% year-over-year—fast but not explosive growth, concentrated among under-35s (16%), with just 1% calling AI their main news source. A key concern for publishers: only 4% of respondents say they often click through from chatbots to original sources, versus 19% for search and 17% for social. The report lands amid a broader shift in which social and video are now the most-used news source globally (54%) while overall trust in news sits at a record-low 37%.

Added: June 17, 2026

Crunchbase's funding roundup for May 30–June 5, 2026 shows AI megarounds multiplying. AI music startup Suno raised $400 million in Series D led by Bond at a $5.4 billion valuation, even as it faces lawsuits from multiple music labels. The same week saw Ramp close $750M (a $44B valuation), Supabase raise $500M (a $10.5B valuation), and brain-inspired AI startup Flourish raise $500M from backers including Jeff Bezos, Lux Capital, and Google Ventures—underscoring that capital is still flooding into AI infrastructure and applications.

Added: June 4, 2026