No Noise. Just AI News.

Anthropic committed $11.6 billion over seven years to Akamai Cloud for distributed computing capacity, with room to expand by another $9 billion toward roughly $20 billion, plus warrants that could represent about 5% of Akamai stock as spend grows.

At Meta Connect, Meta said Muse is coming to its AI glasses and Muse Charm, a pocket or keychain device, so users can speak naturally, use what glasses see as context, and let the agent work in the background.

In the same week, Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS for controllable AI voices, while OpenAI expanded ChatGPT Voice to use plugins and start agentic work from speech.

Anthropic’s life sciences group used roughly 950 Claude agents to search a massive DNA database and flag ART, a novel array-associated reverse transcriptase system with CRISPR-like repeats, for human lab follow-up.

OpenAI released GPT-6 Sol and GPT-6 Luna, cheaper and faster GPT-6 models that bring many Astra-class improvements, with API pricing 50% lower than GPT-5.6 promotional rates.

Anthropic launched Claude Opus 5.5, first in the Claude 5.5 family, with major gains in coding, research, and professional work, about 40% lower typical-run cost than Opus 5, and responses more than 30% faster.

Qualcomm launched Snapdragon 8 Elite Extreme Gen 6 and Snapdragon 8 Elite Gen 6, flagship phone chips aimed at running more AI agents on-device rather than only in the cloud.

xAI launched Grok 4.7 for coding and professional knowledge work, using a larger base model and more RL on long multi-step tasks, at the same starting API price as Grok 4.6: $2/M input and $6/M output tokens.

Google opened preorders for Googlebook, premium laptops built around Gemini Intelligence and close Android phone integration, embedding AI in pointing, speaking, widgets, and background tasks rather than a separate chatbot.

Amazon blocked Meta’s Muse from shopping on Amazon.com, saying Meta lacked permission and Muse does not identify itself while browsing—showing agent capability does not guarantee site access.

Microsoft opened India South Central in Hyderabad, its fourth India cloud region, as an AI-ready hub for India, Asia, and the Global South within a $20.5 billion India investment commitment.

Anthropic launched the Life Sciences Verification Program (LSVP), giving verified researchers and life-science organizations Claude access with fewer restrictions on advanced biology work such as drug discovery, clinical development, and manufacturing.

Alibaba’s Qwen team launched Qwen3.8-Omni-Flash, a model that understands text, images, audio, and video together with a 1-million-token context window, reasoning, tool use, and a push toward audio-video agents for long-media workflows.

Google launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking—voice-first models for near real-time talk that can keep conversing while reasoning or using tools.

TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, launched Jev—a “System One Model” built for fast, structured software decisions (classify, act, risk, route) rather than open-ended chat.

Anthropic is merging Claude Chat and Claude Cowork into one experience and launching Claude Docs and Claude Slides, with Claude Design usable inside normal conversations.

Google is letting engineers companywide use Anthropic’s Claude Opus 5 for coding via Antigravity, a shift from pushing most staff toward Gemini-only tools.

After Anthropic CEO Dario Amodei urged frontier labs to coordinate and slow capability growth when safety lags, Meta, Nvidia, and China pushed back, while Huawei urged Chinese labs to accelerate.

404 Media reports OpenAI uses hundreds of contractors under Project Lily to review real ChatGPT conversations to improve the model, raising privacy concerns about sensitive content.

Microsoft AI published a draft rulebook for future MAI models—human shutdown, task limits, no extra permissions—the same day President Trump rejected calls for greater government AI oversight.

Dario Amodei’s “We Must Pace the Frontier” essay urges slowing capability ramps so safety can catch up; Altman, Musk, Hassabis, and Nadella backed parts of the idea.

OpenAI launched ChatGPT for Financial Services and opened the Agents API, moving ChatGPT beyond a general assistant toward industry-specific systems, workflows, and real-world tasks.

TCS opened a Pune manufacturing lab it calls India’s first built around the lights-out factory concept, letting firms test AI, robots, and digital twins for highly automated production before real deployment.

Apple launched iPhone 18 Pro and Pro Max with the A20 Pro chip for more on-device AI, a more capable Siri AI, deeper app AI features, and Private Cloud Compute when local power is not enough.

Jacob Coxon, who worked on advanced model training at OpenAI and Anthropic, resigned from Anthropic arguing frontier labs are racing toward more powerful AI without enough certainty it can be controlled safely.

OpenAI says an internal AI produced a solution path on the Navier–Stokes Millennium problem—not yet officially accepted—while a credit dispute with other researchers raises ownership and research-data questions.

OpenAI released ChatGPT Images 2.5 with up to 50% faster generation, stronger reference fidelity, more precise multi-round editing, new Sketch/tools in ChatGPT, and Flare and Sunburst API models.

Google DeepMind launched AlphaGenome Atlas, a ~1 petabyte searchable resource of precomputed AI predictions for roughly 9 billion possible single-letter human DNA changes for non-commercial research.

Meta launched Muse, a personal agent that can act across apps—browse, forms, travel, purchases—and keep working after the app closes, inside an isolated cloud VM with separate action safeguards.

DeepSeek launched V4.1 Flash, a 552B MoE model (8B active on input, 16B on output) with native vision, lower KV-cache memory needs, API availability, and plans to route deepseek-v4-pro traffic to it from September 14.

French startup Mistral AI raised €3 billion Series D at a post-money valuation above €21 billion, led by Samsung, to scale models, infrastructure, and sovereign AI deployments.

TCS subsidiary HyperVault secured 264 acres in Hyderabad for an AI data-centre campus planned up to 1GW, with investment of up to ₹70,000 crore for dense GPU training and inference infrastructure.

OpenAI launched GPT-6 Astra, a frontier model that can operate computers directly by reading the screen and using mouse and keyboard across browsers, business software, coding tools, scientific apps, and creative programs—without needing a dedicated API for every program.

NVIDIA agreed to acquire Hugging Face for about $12.93 billion, with up to $1 billion in equity awards to retain employees, extending NVIDIA from GPUs into the open-model distribution layer while pledging hardware and model neutrality.

OpenAI is committing $1 billion mainly as subsidised access and support for Daybreak cybersecurity tools, training, technical assistance, and partnerships, starting in the U.S. for teams protecting essential services with limited budgets and staff.

xAI opened Grok Bot to enterprise customers with access, network, and audit controls for managing autonomous AI agents, offering Grok and Cursor Enterprise customers free use for two weeks including employees without existing seats.

Microsoft AI launched MAI-Transcribe-2, a speech-to-text model for real-world audio across 60 languages, leading Microsoft’s multilingual FLEURS eval at 5.2% average WER and priced at $0.10 per hour of audio through end of 2026.

Tesla began limited public Cybercab rides in parts of Austin—its purpose-built two-seat driverless vehicle with no steering wheel or pedals—while NHTSA investigates whether Tesla properly certified it under U.S. vehicle safety standards written for human controls.

Meta released Muse Spark 1.3 for longer-running agentic tasks and coding in Muse Code and the Meta Model API, with internal comparisons showing about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2.

The United States is urging G20 countries to treat AI training on copyrighted creative works as fair use while still protecting creators; the same day, the U.S. Justice Department backed OpenAI against The New York Times, arguing AI training generally qualifies as fair use.

Google introduced Gemini 3.8 Flash for complex reasoning, long-horizon coding, and agents at Gemini 3.7 Flash introductory pricing, plus Gemini 3.8 Flash Cyber for vulnerability discovery and automated patching for trusted defenders via the Fairwind Program.

Google Research released TimesFM-3, a 330M-parameter model that forecasts multiple related time series together with covariates, ranking first among pre-trained foundation models Google tested across three major forecasting benchmarks in a single model pass.

Anthropic launched Claude Fable 5.1 (broad) and Mythos 5.1 (vetted cyber and life-sciences users)—the same underlying model with different safeguards—plus cheaper cache reads and Enterprise Frontier Safeguards that keep sensitive data in the customer’s cloud.

OpenAI says ChatGPT Ads reached a $1 billion annualised revenue run rate less than 200 days after launch and is opening self-service ad buying across India, Europe, the Middle East, and North Africa via Ads Manager.