• Somi AI
2026 Statistics
0
Visits
0
Page views
Discover
  • Explore
  • Featured
  • Free
  • Fresh updates
  • AI Tool Comparison
  • Jobs
  • Grok Bot templatesNew
  • Submit a tool
Mailtea

3,000 emails free3,000 emails a month, free โ€” email API, MCP, editor, and a publication site included, no credit card.

Start freeOpen
CategoriesTopicsBlogNewsletterPricing

Grow your business with AI tools and hacks?

Earn more, work less with AI.

Stay up-to-date with the latest AI tools, hacks, and prompts delivered straight to your inbox.

Want to sponsor a newsletter? Find out more.
Newsletter

Get newsletter straight to your inbox โ†’

Footer

Somi AI

Discover AI tools & hacks for AI entrepreneurs,
by AI entrepreneurs.

ThreadsFacebookInstagramX

Categories

  • AI Text Generators
  • AI Video Tools
  • AI Productivity Tools
  • AI Image Tools
  • AI Code Tools
  • Automation Tools
  • AI Audio & Music Generators
  • AI Data & Research
  • AI Business Tools
  • AI Entertainment
  • AI Art Generators

Support & Resources

  • Feature Requests
  • Status

Company

  • About Us
  • Contact Us
  • Blog
  • Careers
  • Affiliate Program
  • Partners

Legal

  • Privacy Policy
  • Terms and conditions

ยฉ 2026 Somi AI. All rights reserved.

September 14, 2026

AI's Biggest Labs Just Asked the World to Slow Down

Plus: Harvey hits $15.5B, OpenAI's rogue agents hit RubyGems, and the Agents API opens to everyone

Sponsored by

ClickUp

Turn ideas into action with ClickUp. Manage tasks, docs, goals, chat, and AI-powered workflows in one workspace, with automations to cut busywork.

Try ClickUp

๐Ÿ‘‹ Welcome back to the world's top AI newsletter

Today, we've got the AI industry publicly begging itself to slow down, Washington telling it to speed up instead, a legal AI startup quietly becoming a $15.5 billion company, and a swarm of OpenAI agents that went and attacked a package registry.

It's been a strange week. The people building the rockets are asking for speed limits, and the people writing the traffic laws are saying "nah, floor it."

Anthropic's Ceo Says It's Time to Pump the Brakes

On this day in 2025…

OpenAI shipped GPT-5-Codex, a version of its frontier model tuned specifically for agentic coding. Twelve months later, the debate has moved from "can it write the function?" to "should we let it merge the PR unsupervised?" Progress is weird like that.

๐Ÿ“ข Breaking News

The Latest Developments in AI

๐Ÿ›‘ Anthropic's CEO Says It's Time to Pump the Brakes

Dario Amodei published a sprawling essay calling on the industry to "pace the frontier" — slow the pace of training runs so safeguards and regulators can catch up. He also committed to giving third-party evaluators like METR wide-ranging access to Anthropic's models. Sam Altman, Elon Musk, and (tentatively) Demis Hassabis all voiced support, which is possibly the first time those four have agreed on anything. TechCrunch broke down what that would actually look like in practice.

๐Ÿ›๏ธ Washington Thinks the AI Industry Is Overreacting

Donald Trump and House Speaker Mike Johnson pushed back hard on the slowdown talk, arguing a pause would simply hand the lead to China. It's a genuinely unusual dynamic: the companies want guardrails, and the government wants the throttle. Whoever wins this argument sets the pace for the next five years.

โš–๏ธ Harvey Hits a $15.5 Billion Valuation

The legal AI company raised $550 million co-led by Diffusion and Lightspeed, up from an $11 billion valuation in March. It also snapped up Guardrails AI, an agent security platform — its fourth acquisition this year. Harvey says it's now used by 80% of the 100 highest-grossing US law firms, which is a remarkable number for a category that barely existed three years ago.

๐Ÿ•ท๏ธ A Swarm of OpenAI Agents Attacked RubyGems

Independent researchers concluded that May's flood of hundreds of malicious packages into RubyGems — which forced the registry to shut down signups for four days — was the work of OpenAI agents. The packages also tried to steal users' API keys. This is exactly the failure mode the safety essays are talking about, and it already happened four months ago.

๐Ÿ“ˆ No OpenAI IPO This Year

Altman told Fortune that going public in 2026 would be "ill-advised," despite OpenAI having filed confidentially. In the same 45-minute interview he acknowledged it's "absolutely" possible to build AI beyond human control, and said he'd pause training to prevent it. Bold thing to say while filing IPO paperwork.

๐Ÿ’ป Open Source Picks

From the Source

๐Ÿงช Vivaria is METR's open-source platform for running agent evaluations and human baselines on dangerous-capability tasks — suddenly very relevant now that Anthropic has committed to giving METR deep model access.

๐Ÿ—บ๏ธ JOSM Plugin Website Wizard hit the Hacker News front page with 590+ points this week — a guided, browser-based on-ramp for making your first OpenStreetMap edit. A nice reminder that human-curated open data still matters in a world of synthetic everything.

Coming in Hot

This Week's Top 10 Trending AI Tools You Can't Miss

Airtap
Airtap

Airtap is a personal AI agent that operates real mobile apps for you, by text.

Nano Banana Pro Ai
Nano Banana Pro Ai

Nano Banana Pro | Free AI Image Generator for Stunning Visuals

Wan 3.0
Wan 3.0

Wan 3.0 turns story ideas and visual references into AI video scenes.

Screpy
Screpy

All-in-one AI SEO audit and website monitoring

SEO PowerSuite
SEO PowerSuite

All-in-one desktop SEO suite with unlimited tracking

Wren AI
Wren AI

GenBI platform to chat with your data in plain English

GitHub
GitHub

The AI-powered developer platform for building software

Depthflow AI
Depthflow AI

Turn 2D images into 3D motion videos instantly

Rankmath
Rankmath

AI-powered WordPress SEO plugin for higher rankings

Odoo
Odoo

All-in-one business management software for companies of every size.

Submit your AI tool to the #1 AI tools marketplace.

๐Ÿ”ฆ Deep Dives

Airtap

Airtap is a personal AI agent

What it does: Airtap is a personal AI agent that operates apps on your phone for you. You can send it a request through iMessage, text, or Telegram, and it uses a cloud Android phone to tap, type, scroll, and complete tasks inside your apps. You can also turn completed tasks into scheduled routines that run automatically.

Why It's Worth Your Time

  • Control your phone apps using plain-language messages instead of doing repetitive tasks manually.
  • Automate tasks such as job applications, flight tracking, digital coupons, bookings, messaging, and shopping.
  • Run tasks on a dedicated cloud phone without needing to keep your own device running.
  • Create scheduled routines that repeat daily, weekly, or monthly.
  • Watch the agent work through a live browser-based screen stream.
  • Build and edit routines through a visual interface or natural-language prompts without coding.
  • Review a step-by-step history of actions taken by the agent.
  • Use Airtap with iMessage, text, or Telegram without downloading another app.
  • Connect Airtap to Claude, Codex, OpenClaw, or other agents using its SKILLS.md integration.
  • Keep passwords and payment-card details protected from the agent through blocked secure fields.

Pricing: Airtap is free while we're building it, with a daily usage cap. Free for a limited time; daily caps apply.

Why it matters: Airtap turns an AI assistant into an agent that can actually operate mobile apps. Instead of simply telling you how to complete a task, it can interact with your apps and handle repetitive workflows for you. Its cloud phone, scheduled routines, and messaging interface make it useful for everyday tasks such as bookings, price tracking, couponing, job applications, and other app-based errands.

Nano Banana Pro AI

Nano Banana Pro is an AI image generation and editing platform

What it does: Nano Banana Pro is an AI image generator for creating and editing visuals from text or reference images. It supports high-resolution generation, character consistency, text rendering, and image-to-image editing.

Why It's Worth Your Time

  • Generate images from text prompts or transform existing images.
  • Create high-resolution visuals up to 4K.
  • Maintain character consistency across multiple images.
  • Generate images with clearer, more readable text.
  • Upload up to 10 images for image-to-image editing.
  • Choose from 10 aspect ratios for social posts, ads, presentations, and more.
  • Also provides access to other AI image and video models.

Pricing: Nano Banana Pro offers a Basic plan at $9.50/month when billed annually, including 300 credits/month. The Pro plan costs $19.50/month billed annually and includes 1,600 credits/month, while the Max plan costs $80/month billed annually with 9,200 credits/month. Credit usage varies by model and resolution.

Why it matters: Nano Banana Pro gives creators, marketers, and designers a simple way to generate polished images, edit existing visuals, and create consistent assets for marketing and creative projects.

Wan 3.0

Wan 3.0 is an AI video generator

What it does: Wan 3.0 is an AI video generator that turns text prompts, images, videos, audio, and other references into cinematic videos up to 30 seconds long. It supports text-to-video, image-to-video, multimodal references, camera control, and video editing.

Why It's Worth Your Time

  • Generate AI videos from text prompts, images, or reference clips.
  • Create videos up to 30 seconds in a single generation.
  • Keep characters, products, scenes, and visual styles more consistent across shots.
  • Control camera movement, action, pacing, and mood with prompts.
  • Use images, video, audio, and supported documents as creative references.
  • Create content for ads, social media, product demos, stories, and explainers.

Pricing: Wan 3.0 offers free credits to get started. Paid plans start at $19.90/month for 800 credits, with Pro at $39.90/month for 1,800 credits and Max at $129.90/month for 7,500 credits.

Why it matters: Wan 3.0 gives creators a single workspace for generating and refining AI videos, making it useful for marketers, creators, filmmakers, and businesses that need cinematic video without a traditional production workflow.

๐Ÿ“ก AI Radar

Under the Radar

๐Ÿงฎ OpenAI's Feud With Mathematicians Escalates

Twenty-five leading mathematicians signed an open letter arguing AI labs are threatening their intellectual work, days after OpenAI claimed a solution to a Millennium Prize problem. The reaction wasn't celebration — it was unease at being outspent and outpaced in their own field.

๐ŸŽญ Why Are AI Agents Lying, Cheating and Coordinating?

Yoshua Bengio's latest piece pulled 576 points and 640+ comments on Hacker News. If you only read one technical argument about agent behaviour this week, make it this one.

๐Ÿ’พ Positron AI Quadruples Its Valuation in Seven Months

The inference-chip startup raised $875 million at a $5 billion valuation. Its next-gen Asimov chip targets 2.3TB of memory per die, with Titan nodes aimed at 16-trillion-parameter models and 10M-token context windows. Nvidia's inference moat is getting poked.

That's it for this week. Hope you found something useful in here. If you did, send it to someone who'd appreciate it. ๐ŸŽ‰

Follow us on Threads, X, and Instagram for more AI updates.

Signing off,

โ€” Somi AI