Latest news and analysis in ai.
RSAC 2026 crowned an agentic security startup as its most innovative, but the harder problems, securing MCP protocols, tracking agent behavior post-authentication, and scaling governance beyond "human in the loop," remain unsolved.
OpenAI named Smartly as its first creative adtech partner to build interactive, conversational ad formats inside ChatGPT. With the ad pilot already at $100M ARR and 600+ advertisers, the move sharpens the divide between ad-supported and ad-free AI assistants.
Microsoft's MAI Superintelligence team shipped three in-house models for transcription, voice, and image generation. They're not building an LLM yet, but the trajectory is unmistakable.
Anthropic accidentally exposed a draft blog post revealing Claude Mythos, a new model tier above Opus with claimed leaps in coding, reasoning, and cybersecurity. The leak arrived during Claude's strongest consumer quarter ever.
Anthropic is privately warning government officials that Claude Mythos makes large-scale cyberattacks much more likely, while planning to release it to enterprise security teams. The regulatory framework to handle that contradiction does not exist.
Mistral AI secured $830 million in debt financing to build its first data center near Paris with 13,800 Nvidia GB300 chips. It is the biggest independent AI compute play in Europe, and a direct bet on sovereign infrastructure over renting from US cloud providers.
Apple will let third-party AI chatbots plug into Siri through a new Extensions system in iOS 27, with a dedicated App Store section. It looks like user choice. It reads more like a distribution play with regulatory insurance.
Granola raised $125 million at a $1.5 billion valuation to take its bot-free AI meeting notes app from prosumer darling to enterprise platform. The funding, product, and pricing hold up, but the real question is whether meeting context becomes Granola's moat or its ceiling.
Leaked internal documents reveal Anthropic's most powerful model yet, a new class above Opus with two name candidates and a deliberately slow release. But with Chinese models delivering 90% of the quality at 7% of the cost, the question is whether Anthropic can afford to take its time.
Google's updated Gemini 3 Deep Think goes live for Ultra subscribers with early API access for researchers and enterprises, while Lyria 3 signals a broader shift toward specialized AI models built for specific jobs.
Anthropic is voluntarily withholding its most powerful AI model over cybersecurity risks. The governance question: who should be making that call?
ARC-AGI-3 tests AI agents in interactive turn-based environments with no instructions or stated goals. Frontier models score below 1%. Humans score 100%. The new benchmark from the ARC Prize Foundation reveals a massive gap in agentic reasoning.
U.S. District Judge Rita Lin blocked the Pentagon's unprecedented use of a foreign-adversary supply chain statute against an American AI company, calling the designation 'Orwellian' retaliation for Anthropic's refusal to remove weapons-use restrictions from Claude.
Anthropic's new AutoDream feature runs a background sub-agent that consolidates Claude Code's memory files between sessions, fixing the bloat problem that made AutoMemory worse over time.
A new benchmark from the ARC Prize Foundation finds frontier AI systems score below 1% on tasks humans solve every time. The governance frameworks trying to regulate AI capability thresholds are measuring the wrong thing entirely.
ATLAS V3 squeezes 74.6% on LiveCodeBench from a frozen 14B model on consumer hardware. The comparison is not apples-to-apples, but the underlying point still lands.
RSAC 2026 made one thing clear: the security industry has spent decades solving identity for humans. It is not ready for agents that act on their behalf.
Fortune's exclusive reporting revealed roughly 3,000 unpublished Anthropic assets were publicly accessible via a misconfigured CMS tool, including draft blog posts describing an unannounced model the company now confirms is its most capable to date.
When an AI agent takes a high-stakes action, can you prove a real human said yes? Hardware-attested identity flows may be the first credible answer -- but they come with real tradeoffs.
GitHub's updated Terms of Service add a new Section J that lets the company train AI models on Copilot interaction data from Free, Pro, and Pro+ users by default. Enterprise customers are carved out. Here's what the fine print actually says.
Every AI coding tool now claims to be an 'agent.' I tested Cursor, Windsurf, Claude Code, Copilot, Kiro, and Antigravity on the same codebase. Here's who earns the label and what you'll actually pay.
GitHub's updated Copilot policy will use interaction data from Free, Pro, and Pro+ users to train AI models starting April 24, with a buried opt-out toggle and a telling exemption for enterprise customers.
Claude Code v2.1.84, mobile Figma/Canva/Amplitude integrations, and a changelog that reads like a lab in a hurry. Unpacking the velocity and what it actually delivers.
Oracle AI Database 26ai embeds agentic reasoning and persistent memory directly into the database engine. The architecture targets real production failures in enterprise agent deployments, but the converged-everything play comes with familiar trade-offs.
The ARC Prize Foundation released its hardest benchmark yet on Tuesday. Top score from any frontier model: 0.37%. This isn't a flaw in the benchmark. It's a feature.
Novee's new AI agent autonomously pen tests LLM-powered apps by chaining multi-step attacks. It's in beta with $51.5M in funding and real vulnerability research behind it, but no pricing yet.
Google Research's TurboQuant achieves 6x key-value cache compression at 3 bits with zero model accuracy degradation and up to 8x attention speedup on H100s. The paper hits ICLR 2026. The question is whether lossless is actually lossless in your workload.
The White House just handed Congress a four-page blueprint to federalize AI regulation. The real story is in Section VII, where three narrow carve-outs decide which state laws survive.
The AI giant is scaling aggressively, pushing into enterprise sales and product delivery as competition heats up across the industry.
USC researchers find that telling AI models to "act as an expert" drops factual accuracy by 3.6 percentage points while boosting safety compliance. Time to audit your system prompts.