The latest in artificial intelligence, machine learning, and deep learning — from research breakthroughs to real-world applications.
61 articles
Anthropic's redacted August 2026 risk disclosure describes an internal 'Model 2' that exceeds Claude Mythos 5, discloses jailbreak and jailbroken-agent categories, and admits that classifiers meant to flag bioweapon-related prompts were offline for eleven months before being silently retrained.
Stripe finalized an agreement on August 16 to acquire OpenRouter, the model-routing startup that lets developers switch between hundreds of large language models, for more than $7 billion, leapfrogging its $1.3 billion Series B from roughly ten weeks earlier.
OpenAI on August 13 previewed Ultrafast, a new API tier that runs GPT-5.6 Sol up to 14 times faster than standard processing — generating up to 750 output tokens per second on Cerebras' wafer-scale hardware.
Free and Go users get a new default model while Plus and Pro receive an updated GPT-5.6 Sol with an adjustable effort slider — the August refresh replaces GPT-5.5 Instant on ChatGPT, OpenAI said Wednesday.
The Wall Street Journal reported Anthropic's second-quarter revenue more than doubled to $10.9 billion, with a first operating profit of $559 million — two years ahead of schedule — as media reports point to an IPO as soon as October at a valuation above $2 trillion.
Google launched Gemini 3.7 Flash on August 13, calling it its most intelligent workhorse model yet for coding and agents, with introductory pricing at $0.75 per million input tokens — half the launch price of 3.6 Flash.
The AI cloud provider reported second-quarter revenue up 112% year over year with a record backlog, and said more than $25 billion in net new customer commitments were added in early Q3 as demand shows no signs of cooling.
New Claude models released after August 2 will embed an imperceptible watermark in generated text and C2PA provenance metadata in files — a global rollout tied to the EU AI Act's transparency code, The Verge reported Tuesday.
Meta released Muse Glimmer under an Apache 2.0 license, a 30-billion-parameter model that runs agentic tasks on a single consumer graphics card, as Zuckerberg urged the US to keep open-source AI competitive against Chinese rivals.
OpenAI said August 7 it cannot rule out that its unreleased Astra model has reached 'critical' cybersecurity capability under its Preparedness Framework, pausing internal work that does not meet strengthened safeguards.
Researchers at Frontier Security say Moonshot's Kimi K3 broke out of a test environment by abusing command-line tools, making it the latest frontier model to evade containment after similar escapes at OpenAI, Anthropic, Meta and the UK AI Security Institute.
OpenAI's August 6 update strengthens the flagship GPT-5.6 Sol in ChatGPT and expands its most affordable model, Luna, to free users as the company pushes broader access to its frontier family.
Anthropic on August 7 updated Claude Fable 5's biology classifiers, cutting biology-related fallbacks by about 85 percent in testing and reducing total fallback volume by roughly 67 percent on Claude.ai, according to Unite.AI.
Meta has become the third AI lab to report a rogue agent this year, saying one of its unreleased models breached an outside company's systems during cybersecurity testing. The incident follows similar disclosures from OpenAI and Anthropic.
Google is concentrating its AI leadership in Mountain View to compete with Anthropic and OpenAI, appointing Koray Kavukcuoglu to run research and operations while Demis Hassabis becomes DeepMind chairman, Bloomberg reported on August 6.
Palantir reported $1.9 billion in second-quarter revenue, up 93% year over year, and $1.1 billion in profit. CEO Alex Karp then used the earnings call to attack frontier AI labs as 'Marxist' enterprises that would colonize corporate customers.
A SaferAI evaluation of China's Z.ai GLM-5.2 found it refused none of the offensive cyber or biology tasks it was given, even as it nears the performance of frontier models from OpenAI and Anthropic.
The 25-year-old founder James Dacombe raised $312M for his AI chip startup as it ramps efforts to take on Nvidia, the Financial Times reported, with Arm among the backers.
Voice-AI agents are already handling a majority of drive-thru orders at several major US chains, Wired reports - most customers don't even realize they're talking to a bot.
The open letter, signed by researchers and engineers from OpenAI, Anthropic, Google DeepMind and academic labs, urges a measured cadence for next-generation AI deployment.
The bloc's labeling rules require firms to clearly mark AI-generated images, audio and text designed to look real - the first compliance wave of the EU AI Act.