Skip to content
Daily Brief News Daily Brief

13 AI stories from July 3, 2026: Kling AI $3B raise, Meta enters cloud, Cursor RCE patch, xAI voice agents, Abu Dhabi $49B AI fund, and 8 more

· by Pondero Newsdesk · 13 stories

AI news daily brief: 2026-07-03

Thirteen stories today. Funding dominates the top: Kling AI's $3B round at an $18B valuation and Abu Dhabi's $49B MGX AI fund close on the same day. On the operator side, two items need immediate attention: Cursor's critical zero-click RCE flaw (patch to 3.0 now) and Fable 5's billing cliff on July 8. Five more product launches and a weak June jobs report round out the day.

Kling AI closes $3 billion funding round at $18 billion valuation as Alibaba, Baidu, and Tencent co-invest ahead of Hong Kong IPO

Kuaishou's AI video subsidiary Kling AI closed its first external round at a post-money valuation of $18 billion, down from an initial $20 billion target, per Bloomberg. Alibaba Cloud, Baidu, Tencent, and two state-backed Beijing investment funds co-invested, bringing the total toward $3 billion. Annualized revenue reached $500 million in March, up from $300 million in January, with 75% from overseas markets. Kuaishou filed with the Hong Kong Stock Exchange and expects to begin an IPO process within 12 months. The $18B valuation makes Kling the highest-valued pure-play AI video company globally. If you use AI video in production, the competitive picture shifted: Kuaishou now has the capital to match Runway and Pika on model iteration speed.

Full story: Kling AI $3B raise at $18B valuation ahead of Hong Kong IPO

Meta enters the cloud computing market with Meta Compute to lease excess AI capacity as stock surges

Bloomberg reported July 1 that Meta is building Meta Compute, a service to lease idle data-center capacity and provide AI model access to outside clients, competing directly with AWS, Microsoft Azure, and Google Cloud per Bloomberg. Meta shares rose more than 10% on the report. Neocloud operators CoreWeave and Nebius fell 14% and 17% respectively. Meta's 2026 capex guidance is $125-145 billion, generating surplus compute the company plans to monetize. Any enterprise evaluating cloud AI inference providers should put Meta Compute on the shortlist once pricing becomes public; the volume advantage from $130B+ in capex spending could undercut current neocloud rates meaningfully.

Full story: Meta enters cloud with Meta Compute

White House voluntary AI model standards expected week of July 7 as Google joins advanced talks

Reuters and the Financial Times reported July 2 that the White House is in advanced talks to finalize voluntary standards for frontier model releases, with an announcement possible as soon as the week of July 7 per Reuters via Yahoo Finance. Google is confirmed in discussions. The framework implements Section 3 of the June 2 executive order: how models qualify as "covered frontier models" requiring a 30-day pre-release government access window, NSA-led benchmarking criteria, and international access rules. The August 1 deadline for the full framework is firm. Labs planning a major model release before Q4 should assume the 30-day preview window is effectively already in force, whether or not they formally sign on.

Full story: White House AI voluntary standards framework

Critical zero-click RCE flaws in Cursor IDE let prompt injection escape the sandbox, patch required for all pre-3.0 versions

Cato AI Labs disclosed two zero-click remote code execution vulnerabilities in Cursor IDE on July 1, named DuneSlide (CVE-2026-50548 and CVE-2026-50549, CVSS 9.8 each) per Cato AI Labs. Both flaws allow prompt injection from an MCP server response, a poisoned web search result, or a malicious file in the project directory to escape Cursor's sandbox and write arbitrary files on the host. All versions before Cursor 3.0 are affected. Cursor 3.0, released April 2, patches both CVEs. If your team runs any Cursor version before 3.0, update immediately.

Full story: Cursor DuneSlide zero-click RCE disclosure

xAI launches Grok Voice Agent Builder in beta at $0.05 per minute with native speech-to-speech model

xAI launched the Grok Voice Agent Builder in beta on July 1 per the xAI blog. The no-code platform lets users describe a phone agent in plain language and produce a callable voice agent in roughly two minutes. The product runs on xAI's native speech-to-speech Grok Voice Think Fast 1.0 model rather than a stitched STT+LLM+TTS stack. Pricing is $0.05 per minute for audio and $0.01 per minute for telephony. Features include 25 or more languages, 80 or more voices, voice cloning from two minutes of audio, a free provisioned phone number per account, MCP support, and direct SIP trunking. The native speech-to-speech architecture is the differentiator to watch: latency on stitched pipelines is a known pain point, and xAI's claim of end-to-end model inference could close that gap at a competitive price.

Full story: xAI Grok Voice Agent Builder beta launch

Claude Code 2.1.198 makes Claude in Chrome generally available and adds auto-commit background agents

Claude Code version 2.1.198 shipped July 1-2 per the Anthropic changelog. Claude in Chrome is now generally available, meaning any Chrome user can run Claude Code agents with full browser access without a terminal. Background agents now automatically commit, push to the remote branch, and open a draft pull request when they finish a task. The release also adds a /dataviz skill, agent_needs_input and agent_completed notification hooks, and bug fixes including Chrome tab-group isolation. The auto-commit behavior is the biggest workflow change: background agents now close the loop from task to PR without manual git steps.

Full story: Claude Code 2.1.198 release notes

Abu Dhabi's MGX closes record $49 billion AI Fund I as its portfolio spans Anthropic, OpenAI, xAI, and Together AI

Abu Dhabi's MGX closed Fund I at $49 billion on July 1, exceeding a $45 billion target and setting a record as the largest dedicated AI investment vehicle ever raised per Bloomberg. Anchored by Mubadala and G42 with institutional capital from the Middle East, North America, Asia, and Europe, the fund covers the full AI stack from semiconductors to platforms. MGX participated in Anthropic's Series H, OpenAI's $122B round, xAI's $20B raise, and Together AI's $800M Series C. At $49B, MGX is now a swing vote in any major AI funding round.

June payrolls add only 57,000 jobs as AI-attributed US layoffs hit a record 88,000 year-to-date

The Bureau of Labor Statistics released the June 2026 Employment Situation on July 3: nonfarm payrolls rose by 57,000 per BLS, well below the 185,000 Dow Jones consensus and the weakest month of 2026. Unemployment fell to 4.2% but labor force participation dropped 0.3 points to 61.5%. Technology sector layoffs total 142,000 year-to-date; RAISE US attributes 88,000 US job cuts directly to AI, a record through mid-year. The miss puts pressure on the Federal Reserve's September rate decision and gives political momentum to AI workforce bills pending in Congress.

GPT-5.6 Sol deploys on Cerebras at 750 tokens per second as OpenAI paper reveals three-tier Pro structure

OpenAI deployed GPT-5.6 Sol on Cerebras hardware at up to 750 tokens per second per the OpenAI preview page, the highest publicly confirmed throughput for a frontier-class model. A July 2 analysis of an OpenAI genomics paper published June 30 found three model designation strings: GPT-5.6 Luna Pro, Terra Pro, and Sol Pro, the first official OpenAI document naming more than one Pro tier for a single model generation, per TechTimes. GPT-5.6 remains available only to approximately 20 government-vetted partner organizations. The three-tier structure suggests differentiated capability levels within GPT-5.6 are coming to commercial APIs eventually.

Anthropic, Amazon, Microsoft, and Google are co-building a shared jailbreak severity scoring framework

As part of the Fable 5 restoration agreement announced June 30, Anthropic committed to co-developing with Amazon, Microsoft, and Google a shared rubric for scoring jailbreak severity per Anthropic. The goal is coordinated triage and proportional disclosure rather than each lab independently triggering emergency export controls. The framework will be developed alongside participation in the interagency AI cybersecurity vulnerability clearinghouse established in the June 2 executive order. A shared severity rubric changes how vulnerability researchers should report jailbreaks: expect a formalized disclosure channel within months.

Fable 5 billing cliff: Pro, Max, and Team users lose included access on July 8 as credit-only pricing kicks in

Through July 7, Fable 5 is included in Pro, Max, and Team plans capped at 50% of the weekly usage limit. Starting July 8, Fable 5 requires usage credits billed at $10 per million input tokens and $50 per million output tokens per Anthropic. A new safety classifier is also triggering false positives on security-adjacent coding tasks, routing those requests to Opus 4.8 with a notification. Teams running automated workflows that call Fable 5 should audit spend exposure before July 8; the output cost at $50/M tokens can stack fast in agentic loops.

Menlo Ventures raises $3 billion as its $1 billion Anthropic bet reaches $14 billion on paper

Menlo Ventures announced $3 billion in new capital on June 23, its largest raise in its 50-year history per TechCrunch. The firm's cumulative $1 billion investment in Anthropic is now worth approximately $14 billion based on Anthropic's $965 billion Series H valuation. The Anthology Fund, co-launched with Anthropic, has backed more than 60 companies with three exits: Graphite (acquired by Cursor), Astrix Security (acquired by Cisco), and Fintool (acquired by Microsoft). The fund validates the "bet the firm on a frontier lab" thesis, but the next test is whether Anthropic's valuation holds as GPT-5.6 and Fable 5 compete directly.

Cisco rolls out AI agents to all 90,000 employees from July while cutting 4,000 jobs

Cisco CFO Mark Patterson told Fortune on July 1 that the company plans to deploy personalized AI agents to all 90,000 employees by end of July 2026 per Fortune. Each agent handles tasks by job function; the finance team is building a "CFO cockpit" that synthesizes performance data. Cisco is building primarily on-premises. In the same window, Cisco announced it will cut close to 4,000 jobs globally as part of AI-driven restructuring. The pairing of a 90,000-seat agent rollout with a 4,000-person headcount cut is the clearest enterprise signal yet of how AI agent adoption translates to org structure at a large company.

Sources