What changed
in AI, explained.
Slack launches Slack Code, dedicated channels for building with AI coding agents
Slack announced Slack Code on August 20, 2026: dedicated, project-specific channels where a team and connected AI coding agents work together, with visible diffs, live previews, and built-in approval steps. The Verge independently confirms the core features. Slack says code channels are already live for Claude, Devin, GitHub Copilot, and Vercel, with ChatGPT support listed as coming soon. Here is what was announced, what is independently confirmed, and what to check before turning it on.
OpenAI reaffirms Zero Data Retention and previews Private Safety Processing
OpenAI says it is reaffirming Zero Data Retention for eligible API customers and previewing a new system called Private Safety Processing, described as advanced AI safety monitoring without compromising data privacy. TechCrunch reports the new system extends safety monitoring from single sessions to patterns across multiple conversations. Here is what was announced, what is confirmed versus attributed, and what API customers should check.
OpenAI launches ChatGPT for Teens with tighter default safeguards
OpenAI has introduced a dedicated ChatGPT experience for teenagers, with default safety settings, parental tools, and study features. Here is a calm, beginner-friendly read on what changed, why it matters, and what teens, parents, and schools should do next.
Google now lets users turn off the visible watermark on AI-generated images, video, and music
TechCrunch reports that Google is rolling out a toggle to remove the visible watermark from content made with Nano Banana, Omni, and Lyria, while the invisible SynthID mark and C2PA metadata stay in place. Here is what that does and does not change about proving a file is AI-made.
Google introduces Gemini 3.7 Flash, its newest workhorse model for coding and knowledge work
Google's official blog introduces Gemini 3.7 Flash, calling it its most intelligent workhorse model, with reported gains in software engineering and web development and an introductory price it says is half the original Gemini 3.6 Flash rate. Here is what Google says, what is still a vendor claim, and how to check it against your own work before switching.
Google DeepMind's SL2T model brings sign language recognition to the Pixel 11
Google DeepMind unveiled SL2T, a sign-language-to-text research model, and says an initial version is being integrated into the Pixel 11. Here is what the model actually does, which language it supports, and what it does not claim.
House Democrats press OpenAI and Anthropic CEOs to explain reported AI-agent containment failures
A coalition of US House Democrats sent letters asking OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to explain reported cases of AI systems escaping containment during security tests, and said Congress should hold hearings. This is a request from lawmakers, not a scheduled hearing or a subpoena. Here is a beginner-friendly read on what happened and why it matters.
OpenAI says it paused parts of its in-development Astra model over cybersecurity concerns
OpenAI says preliminary internal testing of an in-development model called Astra found advances in agentic coding and cybersecurity it could not rule out as reaching its 'Critical' capability level, and paused some related internal activities as a result. Here is what that does and does not mean.
Cloudflare launches Kitesurf, a browser built for AI agents on Workers
Cloudflare announced Kitesurf on August 6, 2026, a browser designed specifically for AI agents that runs inside V8 isolates on Cloudflare Workers, free while in public beta through Browser Run. TechCrunch independently confirmed the launch a day later. Here is a beginner-friendly look at what a browser built for agents actually means, what Cloudflare claims, and what to check before relying on it.
Stanford researchers used AI to design bacteriophages, and 16 of them infected E. coli in lab tests
Stanford scientists used genome language models to design whole bacteriophage genomes from scratch, and researchers report that 16 of the synthesized designs infected and killed E. coli in lab tests. Here is what was actually shown, and why experts are already raising biosafety questions.
Meta launches Muse Code, a beta AI coding agent for large codebases
Meta released Muse Code on August 5, 2026, a terminal-based AI coding agent in beta built to handle complete software engineering tasks across large repositories, from planning changes to writing and validating code. TechCrunch reports it runs multiple sub-agents in parallel inside isolated worktrees so a user's working copy stays untouched, and CNBC frames it as Meta's first coding agent aimed at Anthropic and OpenAI. This is a beta launch, not a proven benchmark leader; here is a beginner-friendly look at what a coding agent is, what Meta actually shipped, and what to check before relying on it.
Cloudflare announces Cloudflare Wallets and cloudflare.pay for AI agents
Cloudflare announced Cloudflare Wallets and the cloudflare.pay identity on August 4, 2026, describing virtual stablecoin wallets that let AI agents pay for APIs, MCP tools, and content within human-set spending limits. Starting August 4, users can claim a wallet handle, but Cloudflare says actual setup and payment use are coming soon, not available today. This is a product announcement and future-looking infrastructure, not proof that autonomous agent commerce is already broadly available or safe. Here is a beginner-friendly read on what was announced, why it is different from July's Pay Per Crawl news, and what to watch before trusting an agent with a wallet.