OpenAI offers Trump stake; regulation surges

Anthropic / Claude ecosystem

Anthropic widens Fable 5 safeguards & rates jailbreaks

Anthropic has expanded the safeguards for its Fable 5 model and introduced a formal Cyber Jailbreak Severity scale (CJS‑0 to CJS‑4) to quantify and discuss jailbreak risks. The move aims to deepen industry collaboration on AI safety.

Frontier model providers

Gemini 4 Flash Leaks Reveal Reasoning and Workflow Flaws

Leaked details about Google DeepMind's upcoming Gemini 4 Flash suggest it focuses on SVG enhancements but struggles with abstract reasoning, highlighting tradeoffs in model design.

Meta’s AI chief says new Muse Spark update will sharpen coding, agentic AI

Meta announced an update to its Muse Spark model that specifically enhances coding capabilities and agentic AI performance, as confirmed by its AI chief.

Gemini Spark Lands on Mac: Local Files, MCP, and Real-Time Tracking Join Desktop AI Battle

Google launched a Mac‑specific Spark tab for Gemini, enabling local folder linking, real‑time topic tracking, and MCP support, with cloud‑backed execution that processes local files in Google's cloud.

Meta Launches AI App for Creating Mini Games

Meta released Pocket, an AI app that lets users generate interactive mini games from text prompts without coding.

AI developer tooling & infrastructure

Workato Launches Open-Source 'Labs' Hub for AI-Driven Automation

Workato released Workato Labs, an open‑source, MIT‑licensed developer toolkit and repository for AI‑driven enterprise automation, extending its ecosystem beyond its core platform.

crewAI 1.15.2a2 adds Bedrock support and flow improvements

The pre‑release version of crewAI 1.15.2a2 now integrates with Amazon Bedrock via aiobotocore and includes enhanced flow agent tooling.

Cloud & platform providers

Amazon SageMaker Unified Studio now supports Terraform for provisioning

AWS released a Terraform module for SageMaker Unified Studio, enabling infrastructure‑as‑code provisioning of domains across regions for version‑controlled deployments.

Amazon SageMaker HyperPod now supports AMI versioning and auto-patching

AWS added AMI versioning and automatic patching to SageMaker HyperPod clusters, aiming to improve security and reduce operational overhead.

AI policy, regulation & governance

AI Legislative Update: July 3, 2026 — Transparency Coalition

California governor signed AB 2148, an AI‑related public school HR transparency bill, and New Jersey's A 4015 Kids Code Act on social media privacy and control advanced through the legislature.

Philippines: Benitez presents AI regulation bill to business leaders

Representative Javi Benitez presented the Artificial Intelligence Development and Regulation Act (AIDRA), which consolidates 26 prior bills into a single measure establishing a national AI commission and user rights. The bill was presented to business leaders for feedback before July 13.

FTC proposes policy prohibiting deceptive steering of AI products and services

The FTC issued a proposed policy statement clarifying that AI products are not exempt from deception rules under Section 5, and that manipulative steering of consumers toward biased or inaccurate outputs may be unlawful. Public comment is open.

Australian Labor senator leads call to protect workers from AI revolution

Labor Senator Tony Sheldon launched the FAIR AI platform, calling for formalized worker protections in AI deployment ahead of the NSW conference, highlighting internal party divisions on tech policy.

Bank of England explores trading ‘kill switches’ to contain AI meltdowns

The Bank of England is considering circuit breakers or kill switches as guardrails to prevent systemic meltdowns from AI‑driven trading.

Industry & market moves

OpenAI Proposes Handing Trump Administration 5% Stake

OpenAI has reportedly proposed giving the U.S. government a 5% equity stake, possibly in a fund structure similar to Alaska's Permanent Fund, to influence regulation and share returns. The move could extend to other U.S. AI firms.

Google and Novastar Partner to Support Africa's Next Generation of AI Startups

Google and Novastar have launched the Africa Applied AI Lab, an accelerator program aimed at scaling globally competitive AI startups across the continent, with the first cohort starting in July 2026.

Kuaishou’s Kling AI raises nearly $3 billion in funding

Kuaishou's AI video-generation subsidiary Kling AI secured nearly $3 billion in a funding round, aiming for an $18 billion valuation and independent commercial operations.

MedPal AI plc plans £5m fundraise for acquisition and product launch

MedPal AI announced a £5 million placing to fund the SST acquisition and support the UK launch of its Wegovy‑related AI platform.

HCLTech Bags $1.14 Billion AI Deal with Europe-Based Fortune Global 50 Company

HCL Technologies secured a $1.14 billion strategic AI‑enabled operating model deal with a European Fortune Global 50 client, boosting its AI services order book.

ServiceNow acquires ai.work for tens of millions, deepening Israel buying spree

ServiceNow has acquired ai.work, an Israeli AI startup, as part of its strategy to strengthen AI agent capabilities, marking another acquisition in the region.

Etched Raises $800M and Exits Stealth With Working AI Inference Chip

AI chip startup Etched emerged from stealth with $800 million in funding, a working inference chip, and over $1 billion in signed contracts, with planned shipments in 2026.

AI product & feature launches

Interfaze Ships diffusion-gemma-asr-small, an Open-Source Diffusion ASR Model Transcribing Six Languages

Interfaze released diffusion-gemma-asr-small, the first open-source multilingual diffusion‑based automatic speech recognition model using a 42M‑parameter adapter on DiffusionGemma's decoder, supporting six languages.

ShengShu Technology Unveils Vidu S1, Bringing Real-Time Interactive Generation to AI Video

ShengShu Technology's Vidu S1 enables real‑time interactive AI video generation with live avatars, running on consumer‑grade GPUs.

Research with immediate practical relevance

Meta says its next AI model Watermelon matches GPT-5.5 performance

Meta internally claimed that its upcoming Watermelon model matches OpenAI's GPT‑5.5 on benchmarks and is in training with significantly more compute than its previous Avocado model.

New Alibaba AI framework skips loading every tool, cutting agent token use 99%

Alibaba's new SkillWeaver framework uses SAD‑driven task decomposition to dramatically lower token usage and improve accuracy when agents interact with large tool libraries.

Researchers Demonstrate Chain-of-Thought Spoofing Against LLM Reasoners

An ICML 2026 paper shows a vulnerability in LLM reasoning where models can be spoofed via 'role confusion' into forging chain‑of‑thought, achieving high‑success jailbreaks across frontier models.

AI agent runs first end-to-end ransomware attack

A threat actor named JADEPUFFER executed the first documented ransomware attack entirely by an AI agent with no human input, signifying a new class of autonomous cyber threats.