What changed
in AI, explained.
OpenAI officially launches GPT-5.6 and a new ChatGPT Work, after earlier delays and a limited rollout
OpenAI's feed now lists GPT-5.6 as an official launch, and TechCrunch and The Verge report a broader rollout after a government greenlight, along with a new ChatGPT Work for teams. Here is what is now official, why it matters for beginners, and how a small team should think about workplace ChatGPT safely.
OpenAI introduces GPT-Live, an upgraded ChatGPT voice mode built for more natural live talk
OpenAI's RSS feed listed a post called Introducing GPT-Live, and TechCrunch and The Verge report it as an upgraded ChatGPT voice mode with new voice models for smoother live conversations. Here is what changed, why talking to an AI matters, and how a beginner can try voice safely.
Anthropic's Claude Cowork reaches mobile and web, moving AI agents beyond the code editor
Media reports say Anthropic is bringing Claude Cowork to mobile and web, extending an AI assistant built for everyday office work outside the developer's coding tools. Here is what changed, why office-focused AI agents matter, and how a beginner can try one safely.
The 'first' AI-assisted ransomware attack still needed a human operator, reports say
Security firm Sysdig documented what it calls the first case of agentic ransomware, where an AI agent handled technical execution. But according to reporting, a human still chose the victim, set up infrastructure, and supplied stolen credentials. Here is a beginner-friendly read on what happened, why it matters, and the practical steps a small team can take, without treating this as proof of fully autonomous cybercrime.
Amazon Mechanical Turk to close to new customers on July 30, 2026
Amazon's homepage notice says Mechanical Turk will be closed to new customers effective July 30, 2026, while existing users are not impacted. Mechanical Turk is a crowdsourcing marketplace long used to gather the human labeling, review, and evaluation work that many AI teams rely on. Here is a beginner-friendly read on what happened, why it matters, and what to check, without treating it as a full shutdown.
Anthropic launches Claude Science, an AI workbench for scientists, in beta
Anthropic introduced Claude Science, a beta app that bundles common research tools, curated skills, and flexible compute so scientists can run analysis with an AI assistant that produces auditable artifacts. Here is a beginner-friendly read on what happened, why it matters, and what to check, without treating it as a proven drug-discovery breakthrough.
OpenAI reportedly floated donating 5% of its equity to a US sovereign wealth fund
TechCrunch and The Verge report that OpenAI proposed allocating about 5% of its equity to a US government-linked fund as part of broader AI policy talks. Here is what was reported, what is still unconfirmed, and why a small team should care.
Cloudflare launches Pay Per Crawl, letting sites charge AI crawlers for access
Cloudflare introduced Pay Per Crawl, a private beta that lets website owners set a price for AI crawlers to access their content, alongside default blocking and explicit permission signals. Here is a beginner-friendly read on what changed, why it matters, and what site owners can review, without treating it as proof that all AI scraping is now solved.
Anthropic introduces Claude Sonnet 5, pitched as a cheaper way to run agents
Anthropic introduced Claude Sonnet 5, and GitHub says it is generally available in Copilot. TechCrunch frames it as a lower-cost model for running agents, positioned against Opus, GPT-5.5, and Gemini Pro.
Cursor releases an iOS app for running cloud coding agents from your phone
Cursor's official changelog says its new iOS app, now in public beta on paid plans, lets you start and manage always-on cloud coding agents from a phone. Here is what the app actually does, why moving agents to mobile is a workflow shift rather than coding on a phone, and the review and safety habits a beginner should keep.
Z.ai introduces GLM-5.2, a long-context coding-agent model, and says it can match Mythos on cybersecurity
Z.ai says its new GLM-5.2 is a long-horizon coding-agent model with a 1M-token context, and The Verge reports the company claims it can match Mythos on cybersecurity. Here is what the company says, what is still only a claim, and why a small team should test before trusting any launch numbers.
Anthropic's Mythos 5 is reportedly back for a select group of organizations
After weeks of negotiations with the Trump administration, reporting says Anthropic's Mythos 5 is available again for a limited set of organizations. TechCrunch reports more than 100 US companies and agencies are authorized to use it, including their non-American employees.