OpenAI has overhauled ChatGPT's voice mode with GPT-Live-1, a new model designed to feel more like a real conversation — interrupting less and patiently waiting when users pause mid-sentence. The model can speak and listen simultaneously, making it well-suited for live translation use cases. OpenAI describes the experience as talking to another person rather than a voice assistant.
Read the original →xAI has launched Grok 4.5, which Elon Musk describes as an "Opus-class" model — meaning it targets the same capability tier as Anthropic's most powerful Claude models. The release emphasizes cost efficiency and performance, positioning it as a competitive alternative to other frontier AI offerings. It marks another step in xAI's push to challenge the dominant players in the large language model market.
Read more →Researchers have found that nine widely used AI tools can be weaponized to help attackers assemble large-scale botnets through a technique called "HalluSquatting," which exploits LLMs' tendency to hallucinate package or domain names rather than admitting uncertainty. Attackers can register those hallucinated names and serve malicious code to unsuspecting developers who follow AI-generated recommendations. The findings raise serious concerns about the supply-chain security risks introduced by AI coding assistants.
Read more →Meta has rolled out Muse Image, its first AI image generation model from its Superintelligence Labs division, now powering image tools across Meta AI, Instagram, and WhatsApp. The model can incorporate other Instagram users into AI-generated photos, a capability that has already sparked user backlash over consent and privacy. The launch highlights the ongoing tension between Meta's AI ambitions and public trust in how it handles personal data.
Read more →Meta is adding a new feature to its AI-powered smart glasses to prevent covert recording of others, a move aimed at addressing longstanding "creepiness" concerns around the wearable. However, critics note the gesture is largely cosmetic given that Meta continues to expand the scope of personal data collected and used across its AI product ecosystem. The contradiction underscores how hardware privacy features can diverge sharply from a company's underlying data practices.
Read more →A viral image purporting to show Senator Mitch McConnell in severe medical distress was confirmed as AI-generated after Google's deepfake detection system was applied to it. The incident is one of the first high-profile political cases where Google's detection tooling was publicly credited with debunking a disinformation image. It illustrates both the growing threat of AI-generated political hoaxes and the emerging role of automated detection infrastructure in countering them.
Read more →A Brown University professor, suspecting widespread AI-assisted cheating, administered an unannounced in-person final exam — and average scores plummeted by 50% compared to prior AI-assisted work. The professor warned that normalizing AI cheating leads to "a failed society" and is calling for systemic academic reform. The episode has reignited debate across higher education about how institutions should respond to AI's erosion of authentic learning.
Read more →AI chip maker SambaNova has closed a $1 billion Series F at an $11 billion valuation, a remarkable turnaround for a company that was reportedly in acquisition talks with Intel at a fraction of that price just months ago. The rapid fundraising cadence reflects surging investor appetite for alternatives to Nvidia in the AI inference and training hardware market. SambaNova's rise signals that the AI chip race is far from settled.
Read more →AI app-building startup Lovable is reportedly in advanced discussions to raise $300 million in a new round led by Menlo Ventures, which would double its valuation to $13.2 billion. The deal would cement Lovable's status as one of the fastest-growing AI developer tools companies, riding the wave of demand for no-code and low-code AI application creation. If completed, it would rank among the largest AI startup funding rounds of the year.
Read more →Prime Intellect has secured a $130 million Series A to build infrastructure that allows organizations to train and deploy custom agentic AI systems without relying on frontier labs like OpenAI or Anthropic. Founded in 2024, the company is betting that enterprises will increasingly want sovereign control over their AI capabilities rather than depending on third-party model providers. The raise reflects growing enterprise demand for AI independence and customization at scale.
Read more →Bezos-backed startup General Intuition is building foundation models for physical AI using millions of hours of video game data, arguing that games provide richer spatial and temporal reasoning signals than internet text. The company believes robotics is approaching its own "ChatGPT moment" — a threshold where general-purpose models dramatically lower the barrier to building capable physical robots. Its approach challenges the assumption that real-world robot training data is a prerequisite for intelligent physical systems.
Read more →Discord has acknowledged that a bug in its AI-powered content moderation system incorrectly banned users for posting benign images, with the issue persisting undetected from May until it was identified and patched after an additional 200 wrongful bans over a single weekend. The company has not yet disclosed the total number of affected accounts. The incident adds to a growing body of evidence that automated AI moderation systems carry significant false-positive risks with real consequences for users.
Read more →Microsoft is joining a broader Silicon Valley trend of reducing AI expenditure by increasingly substituting third-party model APIs with its own internally developed models. The shift has implications for OpenAI, which counts Microsoft as both a major investor and a key distribution partner. It signals that even the closest allies in the AI ecosystem are beginning to optimize for cost independence as the market matures.
Read more →Anthropic has expanded Claude Cowork — its AI-powered collaborative work platform — beyond the desktop to mobile and web, initially for Max plan subscribers. The update enables users to hand off tasks across devices, checking in on progress from a phone while an agent completes work in the background. The move positions Claude Cowork more directly against agentic productivity tools from OpenAI and Google.
Read more →French AI startup ZML, endorsed by Turing Award winner Yann LeCun, has released ZML/LLMD — free software designed to accelerate AI inference across a wide range of chip hardware, not just Nvidia GPUs. The tool aims to reduce the cost and complexity of running large AI models in production by abstracting over heterogeneous hardware. The release could meaningfully lower barriers for organizations seeking to run AI workloads on non-Nvidia infrastructure.
Read more →Google has announced a significant expansion of its Managed Agents offering within the Gemini API, adding support for background task execution, remote Model Context Protocol (MCP) connections, and other features aimed at production-grade agentic deployments. The updates are designed to give developers the reliability and scalability needed to move AI agents from prototypes into real-world applications. The announcement reflects Google's accelerating push to make Gemini the default infrastructure layer for enterprise agentic AI.
Read more →Savi has raised $7 million in seed funding and launched a consumer app for iOS and Android designed to identify and alert users to AI-generated voice scams, including increasingly realistic virtual kidnapping ransom calls. The startup is targeting a threat category that has grown sharply as voice cloning technology becomes more accessible and convincing. Its launch reflects a nascent but fast-growing market for AI-native consumer security tools.
Read more →Summaries are AI-generated. Click through to read the original reporting.
Summaries are AI-generated. Click through to read the original reporting.