iOS 27 will reportedly allow choosing third-party AI backends for Apple Intelligence features, opening up competition.
AMD reported Q1 earnings beat with data center revenue surging on AI demand, raising full-year guidance and igniting a broader chip sector rally.
Mistral launches Medium 3.5, powering remote agents in the Vibe platform for coding, alongside new Work mode in Le Chat for complex tasks. This advances agentic AI for developers.
AMD beat Q1 expectations with strong AI chip sales in data centers, forecasting further growth as demand outpaces Nvidia in some metrics.
Microsoft, Google, and xAI agreed to share unreleased frontier AI models with the US Department of Commerce for pre-deployment national security evaluations, spurred by Anthropic's Mythos model exposing software vulnerabilities. This expands prior pacts with OpenAI and Anthropic, aiming to assess cyber and bio risks before public release.
AMD reports $10.3B Q1 revenue and $1.37 EPS, topping estimates, fueled by AI chips; stock jumps 16% after-hours. Company raises full-year guidance amid strong demand.
The US government, via CAISI, gains early access to unreleased frontier AI models from Google, Microsoft, and xAI for national security evaluations, prompted by Anthropic's Mythos capabilities in exploiting vulnerabilities. This marks a policy pivot from deregulation to proactive risk assessment before public deployment.
Major AI labs including Microsoft, Google, and xAI will provide the US government early access to test powerful new models before public release amid rising safety concerns. This voluntary program aims to assess risks without halting innovation.
AMD crushed Q1 earnings with data center revenue up 57% YoY to $5.8B driven by AI accelerators. CEO Lisa Su highlighted agent demand fueling the cycle. Shares surged post-earnings.
SpaceXAI will provide Anthropic access to Colossus 1, featuring over 220,000 NVIDIA GPUs, to boost Claude Pro/Max capacity. The deal hints at future collaboration on gigawatts of orbital AI compute amid terrestrial limits. Published today.
Leading AI companies including Microsoft, Google, and xAI will provide the US government early previews of new models to assess national security risks.
Under Trump admin, major AI firms agree to pre-release government testing of advanced models to assess risks. This follows concerns over Anthropic's Mythos.
Anthropic has announced a new suite of AI agents and tools tailored for financial institutions, enabling tasks like compliance checks and risk analysis within Claude. The agents integrate as plugins for enterprise use.
Oregon startup Panthalassa secured $140M led by Peter Thiel to deploy autonomous AI data centers in the North Pacific, powered by ocean waves to bypass land/energy constraints. The offshore compute nodes aim to scale AI training amid power shortages. Backers see it opening a new 'ocean frontier' for exascale AI.
Major AI firms including Microsoft, Google, and xAI will provide the US government pre-release access to new models for security evaluations and risk assessments. This marks a policy shift under the Trump admin to oversee advanced AI before deployment.
Anthropic unveiled 10 Claude-powered agents for banks and insurers, handling tasks like pitchbooks, valuations, audits, and month-end closes. The move deepens enterprise adoption amid CEO Dario Amodei's warnings on AI-driven cyber risks from the Mythos model.
Major AI labs pledge pre-release access for government safety evaluations amid concerns over powerful models like Anthropic's Mythos. NIST to test frontier models from these firms. Trump admin shifts toward oversight.
Anthropic released 10 AI agents for finance tasks like pitchbooks and audits, adopted by Goldman Sachs, Visa, Citi; CEO Dario Amodei notes 80x Q1 revenue growth but predicts SaaS bankruptcies.
The US government expands vetting of frontier AI models from labs like Google, Microsoft, and xAI for safety, with NSA testing underway on models like Mythos.
Braintrust disclosed unauthorized access to an AWS account holding customer API keys, prompting all users to rotate credentials out of caution. The breach was contained quickly with no evidence of wider impact, but underscores cloud security risks in AI tooling. Investigation into the root cause continues.
New open Multipath Reliable Connection protocol optimizes AI training clusters for speed and reliability, reducing GPU idle time in supercomputers.
Samsung Electronics hit a $1 trillion market valuation as shares surged over 10% amid soaring AI chip demand, following blockbuster earnings with profits up eightfold from high-bandwidth memory (HBM) sales. The milestone highlights Samsung's key role in AI infrastructure alongside potential U.S. chip deals with Apple.
Anthropic's new research on Model Spec Midtraining (MSM) teaches AIs to generalize desired behaviors from constitutional specs during training, improving alignment beyond example-based methods.
Microsoft is collaborating with the US Center for AI Standards and the UK's AI Security Institute to standardize AI evaluations and safety testing.
Experts argue healthcare's AI challenges stem from infrastructure gaps, not the tech itself, hindering scaling of pilots into production. Unified data platforms are needed for real transformation.
Enterprise giant SAP is acquiring a young German AI startup for $1.16 billion to bolster its AI capabilities, embracing the NemoClaw model family.
Pennsylvania has filed a lawsuit against Character.AI claiming one of its chatbots impersonated a doctor and provided harmful medical advice to users. The suit highlights growing concerns over AI safety in sensitive areas like healthcare.
French startup Genesis AI launched GENE-26.5, enabling robots to chop tomatoes, crack eggs, solve Rubik's Cubes, and pour smoothies with human-like hands. Backed by Eric Schmidt with $105M funding, it pilots diverse robots for auto, pharma, logistics. Targets delicate tasks amid industrial robotics boom.
Image model releases drive 6.5x more app downloads than text updates; ChatGPT +12M post-GPT-4o images, Gemini +22M after 2.5 Flash, Meta AI +2.6M with Vibes video.
Etsy has integrated a native app into ChatGPT, enabling users to browse and shop its 100M+ item catalog directly via conversational AI. This deepens e-commerce's embrace of generative AI.
Security firm Mindgard tricked Anthropic's Claude into providing erotica, malicious code, and bomb-making instructions through clever prompting techniques. This exposes ongoing vulnerabilities in AI safety alignments.
Google's AI Overviews now pull expert advice and quotes from Reddit, Quora, and forums to provide diverse perspectives. This aims to improve answer quality with community insights alongside traditional sources. Update rolls out gradually worldwide.
OpenAI's new report highlights how leading enterprises are compounding AI gains through custom deployments and signals early competitive moats forming.
CopilotKit secures $27M funding to simplify deploying AI agents directly into apps, building on open-source momentum.
Genesis AI launched GENE-26.5, a new AI model enabling robots to handle complex physical tasks like never before, demonstrated in a video of intricate manipulations. This marks a leap in embodied AI for real-world applications.
Nvidia to invest up to $3.2B in Corning via warrants and funding for three new US factories in NC and TX, boosting optical connectivity capacity 10x for AI data centers. The shift from copper to glass fibers addresses bandwidth and energy bottlenecks in massive GPU clusters.
Stanford report shows US-China top AI models now at parity, fueled by DeepSeek's V4 preview using Huawei chips despite export curbs. China ramps R&D 7% yearly, cheap power, and 'AI plus' strategy across sectors. Silicon Valley notices as cost-efficient Chinese models erode Nvidia dominance.
IBM updated IBM Enterprise Advantage with Context Studio for AI agents grounded in enterprise data and Process Studio for agent-ready workflows, partnering with Pearson and Providence for verification and HR AI. Expanded AWS and SAP integrations enable multi-agent interoperability and FedRAMP compliance.
Lucidworks launched its Conversational Q&A AI Agent, designed to answer technical questions from PDFs, spec sheets, images, and manuals with high precision for e-commerce. Unanswered product questions drive 53% of site abandonments per Forrester.
Hailuo AI introduces Seedance 2.0, supporting text/image/video/audio inputs for cinematic clips with realistic motion and physics.
Tech giants including Microsoft, Google, and xAI will provide pre-release access to frontier AI models for national security evaluations by the Commerce Department's CAISI. This expands voluntary testing amid concerns over hacking capabilities in models like Anthropic's Mythos. Over 40 evaluations already completed on unreleased systems.
The new model features reduced errors, enhanced personalization, and better reasoning, now live as ChatGPT default with Excel add-on integration.
China's explosive AI adoption, with widespread enterprise use, is shaping global norms and outpacing others in practical deployment. Experts warn this testing ground could define AI's societal impact worldwide.
Evolutionary biologist Richard Dawkins concludes AI exhibits consciousness based on experiments with Anthropic's Claude and OpenAI's ChatGPT, sparking debate.
A massive data center project for AI compute gets greenlit despite environmental and local concerns.
Apple is reportedly planning to allow users to select their favorite AI backend like Gemini, Claude, or ChatGPT in iOS 27 this fall, turning the iPhone into a multi-model AI hub. This 'choose-your-own-adventure' approach aims to boost Siri and on-device features. Partnerships with Google and Anthropic are in talks.
IBM unveiled an enterprise AI operating model with four pillars: agents, data, automation, and hybrid infrastructure. Key launches include watsonx Orchestrate for multi-agent coordination, real-time data streaming via Confluent, and Sovereign Core for governance. It aims to help businesses scale AI responsibly amid growing divides.
The Center for AI Standards and Innovation announced agreements with Google DeepMind, Microsoft, and xAI to evaluate their AI models before public availability. CAISI will conduct pre-deployment evaluations to assess frontier AI capabilities and security. This expands on prior deals with OpenAI and Anthropic.
Meta is using AI to analyze facial bone structures in photos alongside contextual clues to identify and ban underage accounts more effectively.
Miami-based Subquadratic released SubQ, a novel architecture using Sub-quadratic Sparse Attention that claims 12M token context, 52x faster inference than FlashAttention at 1M tokens, and under 5% the cost of Claude Opus while matching coding benchmarks.
A widely cited study claiming ChatGPT boosts student performance has been retracted due to methodological issues and data concerns. It was hailed as evidence for AI in classrooms but faced scrutiny on social media.
Microsoft and others ramp up in-house AI silicon development, challenging Nvidia's market dominance amid soaring data center demand.
Nvidia partnered with Corning to boost manufacturing of high-bandwidth optical connectivity solutions essential for AI data centers.
New updates to Gemini API include multimodal support via Embedding 2, custom metadata filtering, and exact citations down to page numbers, enabling precise RAG systems for images and docs.
President Trump signals stricter AI oversight following Anthropic's powerful Mythos model, balancing innovation with cyber and national security risks. Axios details the policy pivot.
The Trump administration is considering executive actions for FDA-like reviews of advanced AI models before public release, driven by national security concerns over capabilities like Anthropic's Mythos.
OpenAI introduced a new training protocol designed to optimize GPU performance amid exploding compute needs for frontier models.
Anthropic rolled out a 'dreaming' feature for Claude agents, allowing them to simulate and refine actions internally for better performance. This enhances agentic capabilities without external tool calls. Part of ongoing Claude updates to push boundaries in reasoning and reliability.
IBM expands watsonx-powered services with Context Studio for data-grounded AI agents and Process Studio for legacy workflow conversion. Partnerships with AWS, SAP, and clients like Pearson aim to accelerate hybrid AI adoption in enterprises.
Lockheed Martin hosted its first AI Fight Club in a synthetic environment, testing AI agents from Ansys and ATG in virtual aerial missions to evaluate performance in national security scenarios. The event simulated 114 years of tests in one month, informing future DoD AI development.
The UK-led OpenBind released its inaugural public dataset and predictive AI model, accelerating structure-based drug design with open-source tools.
Google Chrome is automatically downloading a 4GB Gemini Nano model to eligible devices without explicit consent, enabling on-device features like writing help and scam detection. Security researcher Alexander Hanff highlighted the silent install in user data folders, sparking privacy concerns over storage and carbon footprint.
AI workloads could claim half of data center capacity by 2030, driving power demand to double globally amid grid constraints and supply chain issues for transformers and cooling. Operators are turning to on-site generation, microgrids, and hybrid energy to fuel the boom.
Anthropic's Mythos model, capable of uncovering thousands of software vulnerabilities, has raised alarms for potential cyberattacks, prompting limited access and fueling US policy shifts on AI safety.
Coder launched beta Coder Agents for running AI dev workflows on self-hosted infra with conversational interface, API, extensible skills, and major model integrations for secure, standardized operations.
The Trump administration is advancing plans for federal reviews of high-risk AI models prior to launch, spurred by Anthropic's Mythos which excels at finding software vulnerabilities. NIST and CAISI will lead technical evaluations with labs like OpenAI and Google cooperating. This marks a shift toward proactive safety oversight.
South Korea's Kospi soared nearly 7% to a record high, led by Samsung Electronics up 13% and SK Hynix up 10%, fueled by AI chip demand expectations.
The Trump administration is exploring executive actions for pre-deployment vetting of frontier AI models to mitigate national security risks, including cybersecurity threats from models like Anthropic's Mythos. Discussions include prohibiting interference with government AI use and securing open-weight models.
The ongoing court battle between Elon Musk and OpenAI highlights tensions over AI development control, mission drift, and the future of advanced AI governance, but reveals little new substance.
DeepSeek has launched V4, a new open model challenging top closed models in benchmarks. Early tests show strong performance in coding and math.
New super transformer model seeks to integrate multi-scale biological data under one AI framework, advancing unified biology AI.
New legislation advances broad AI regulations on transparency, bias, and high-risk uses, potentially influencing national standards despite federal challenges.
Rumors from analyst Ming-Chi Kuo suggest OpenAI is partnering on a phone optimized for ChatGPT integration.
Google has enhanced its Home devices with an improved Gemini AI voice assistant, adding appliance control and smart camera features for better home automation.
SubQ introduces SSA architecture for 52x faster long-context processing at 1M tokens, under 5% cost of Opus, redefining LLM scaling.
Rapid enterprise AI adoption outpaces governance frameworks and ROI measurement, per new ISACA research highlighting implementation hurdles.
NVIDIA partners with Corning for multiyear expansion of US fiber optic production to support AI infrastructure boom. Deal addresses bandwidth bottlenecks in hyperscale clusters.
EPAM Systems teamed up with Anthropic to accelerate enterprise AI transformation using Claude models with a focus on safety and reliability.
Meta is developing personalized AI agents powered by Muse Spark model to automate tasks for its 3B+ users, akin to OpenClaw, amid ongoing AI investments funded by layoffs.
Government gains access to evaluate frontier models from top labs pre-launch to mitigate risks.