★ Top Story
The VergeOpenAI has received the Trump administration's approval to publicly release GPT-5.6, which CEO Sam Altman called "the best model we have ever produced." The launch is accompanied by a broader product push called ChatGPT Work, which includes a rebranded Codex tool capable of running autonomous workflows for hours at a time. The new model family promises improvements across cybersecurity and enterprise productivity, and will serve as the preferred engine for Microsoft Copilot 365.
Read the original →
TechCrunch AIOpenAI's second-in-command, Fidji Simo, is departing her full-time role as AGI chief after a medical leave for a neuroimmune condition stretched far longer than anticipated. She will transition to a part-time advisory role, creating a significant leadership gap at a critical moment as OpenAI eyes a potential IPO and battles Anthropic for enterprise dominance. The timing adds internal uncertainty to an already turbulent week of product launches and legal challenges.
Read more →
TechCrunch AIThe New York Times and allied news publishers have filed a motion for sanctions against OpenAI, alleging the company concealed internal tools and datasets that could identify copyrighted journalism reproduced in ChatGPT outputs. Ars Technica separately reports that OpenAI may have faked an inability to search its own training data and deleted billions of ChatGPT logs. If proven, the misconduct could constitute a fatal misstep in OpenAI's legal defense against the publisher coalition.
Read more →
MIT Technology ReviewAnthropic researchers have developed a new interpretability tool called the Jacobian lens that offers the clearest window yet into what large language models are actually doing as they reason through problems. The findings range from expected to unsettling, revealing that Claude appears to work through concepts in a latent internal space before producing outputs. The breakthrough could have significant implications for AI safety research and our understanding of emergent model behavior.
Read more →
TechCrunch AIElon Musk is publicly courting Anthropic to host its models on his infrastructure, pledging that he will not interfere with or cut off access to the company's services. The assurance comes as Anthropic weighs a partnership worth an estimated $40 billion in potential revenue, a decision complicated by Musk's adversarial history with AI rivals. Musk also praised AI game studio Mythos/Fable during the same public remarks.
Read more →
TechCrunch AIMeta has launched Muse Spark 1.1, a significant upgrade to its in-house AI model, and opened it to developers via the new Meta Model API for integration with coding tools. The company is positioning Spark around enterprise-scale agentic workloads — including large code migrations and automated bug fixing — where it hopes to differentiate from GitHub Copilot and Cursor. Meta describes the new model as a "step-change" improvement over the first generation released in April.
Read more →
TechCrunch AIA new analysis finds that the anticipated public offerings of Anthropic, OpenAI, and SpaceX are collectively projected to generate more value than all U.S. venture-backed exits combined since the year 2000. The staggering figures underscore just how concentrated the current technology boom has become around a handful of AI and space companies. The comparison raises fresh questions about market concentration and whether the broader startup ecosystem can sustain comparable returns.
Read more →
TechCrunch AIOpenAI is sunsetting Atlas, its standalone AI-powered browser, less than a year after its October launch. Rather than abandoning agentic browsing entirely, the company is migrating key features into its desktop application and a new Chrome extension as part of the ChatGPT Work rollout. The pivot suggests OpenAI is consolidating its product surface rather than maintaining a fragmented portfolio of standalone tools.
Read more →
TechCrunch AIEnterprise AI agent builder Lyzr closed a $100 million funding round that was orchestrated end-to-end by one of its own AI agents — handling investor outreach, scheduling, and deal logistics autonomously. The stunt serves as a live product demonstration, signaling growing confidence among founders that agentic AI is mature enough for high-stakes business processes. The raise also reflects continued investor appetite for enterprise automation plays despite broader market caution.
Read more →
The VergeMicrosoft's 2026 sustainability report reveals the company's carbon emissions climbed 25 percent last year to 34 million metric tons, a direct consequence of its rapid AI infrastructure buildout. The figure puts Microsoft further behind its own stated climate goals and reignites the debate over the environmental cost of the AI boom. The disclosure is likely to intensify scrutiny of hyperscalers' sustainability commitments as data center construction accelerates globally.
Read more →
TechCrunch AIThe process by which the Trump administration evaluated and approved GPT-5.6 for public release remains largely opaque, raising concerns about the rigor and transparency of frontier model safety reviews. Experts note that the exact nature of the dialogue between regulators and AI labs like OpenAI and Anthropic has not been publicly disclosed. The ambiguity is fueling calls for clearer, codified standards governing how governments assess the safety of powerful AI systems before release.
Read more →
TechCrunch AIGoogle is expanding its AI disclosure requirements to cover all ads that use synthetic or AI-altered content, not just political advertising as was previously the case. Users will be able to see a "created or edited with AI" label via Google's My Ad Center across Search, Discover, and YouTube. The move comes amid growing regulatory and public pressure for greater transparency around AI-generated media in commercial contexts.
Read more →
Ars TechnicaIn a landmark preclinical trial, surgeons successfully directed humanoid robots to perform live surgical procedures on pigs, marking the first time general-purpose humanoid platforms have been used in an operating theater setting. The research is testing whether humanoid form factors — rather than purpose-built surgical robots — can meet the precision and adaptability demands of real surgery. A successful path to human trials could dramatically expand the addressable market for humanoid robotics.
Read more →
TechCrunch AIOllama, the open-source tool that lets developers run large language models locally on their own hardware, has raised $65 million and grown to nearly 9 million users with over 176,000 GitHub stars. The Benchmark-backed company has become a go-to solution for developers who want to experiment with AI without relying on cloud APIs or sending data to third parties. The funding will help Ollama scale infrastructure and expand its model library as local AI inference gains mainstream traction.
Read more →
TechCrunch AIMeta has confirmed that its next generation of in-house AI accelerators will enter mass production in September, built around a modular architecture designed to adapt as AI workloads evolve. The chips are part of Meta's broader strategy to reduce dependence on Nvidia and control more of its AI infrastructure stack. A successful ramp could meaningfully cut Meta's compute costs and give it greater flexibility in deploying future model generations.
Read more →
The VergeMicrosoft announced it is now deploying AI internally to identify security vulnerabilities earlier in the development cycle, a change it says will result in a higher volume of fixes shipped in each monthly Patch Tuesday release. The move comes as AI-assisted hacking tools have lowered the barrier for attackers, making faster defensive patching cycles increasingly critical. The announcement signals a broader shift toward AI-augmented security operations across the enterprise software industry.
Read more →
Ars TechnicaA Brown University professor who suspected widespread AI-assisted cheating required students to sit an in-person final exam, and average scores plummeted by roughly 50 percent compared to prior AI-assisted work. The dramatic gap has reignited a fierce debate on campus and beyond about academic integrity, the hollowing out of genuine learning, and what universities should do in response. The professor warned that normalizing AI dependency in education risks producing "a failed society."
Read more →
TechCrunch AIMeta's Muse Image tool allows any user to generate AI images by tagging public Instagram accounts, effectively using those accounts' photos as style or subject references without explicit consent. TechCrunch has published a guide explaining how Instagram users can adjust their settings to prevent their public photos from being used as AI generation inputs. The feature has drawn criticism from creators and privacy advocates who argue that opt-out — rather than opt-in — consent is insufficient.
Read more →No stories from this source today.