Top stories in artificial intelligence — models, tools, and enterprise deployment.
Anthropic launched Claude Opus 4.7, improving software engineering performance on complex long-running coding tasks and bumping vision capabilities to handle higher-resolution images. The model is available on all Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at the same pricing as Opus 4.6 ($5 / $25 per million tokens in/out).
Alongside the model, Anthropic shipped two new products: Claude Design — a Labs product for generating visual outputs like prototypes, slides, and one-pagers — and Claude Security, a public beta for Enterprise customers that adds code vulnerability scanning, proposed fixes, scheduled scans, and workflow integrations powered by Opus 4.7.
Anthropic also announced a compute deal with SpaceX, gaining access to the full capacity of the Colossus 1 data center — over 220,000 NVIDIA GPUs, 300+ megawatts — which will directly benefit Claude Pro and Max subscribers.
anthropic.comAnthropic's unreleased Mythos model, described as "far ahead" of existing models in cybersecurity capabilities, has drawn attention from governments, banks, and utility companies. Anthropic says it's not comfortable releasing it publicly and is restricting access to a select group of approved organizations only.
The situation echoes broader industry conversations about the dual-use risks of frontier AI — particularly models capable enough to substantially assist in offensive cyber operations. Mythos has reportedly contributed to Anthropic's $200B+ commitment to cloud infrastructure in partnership with Google Cloud.
GPT-5.5 Instant began rolling out on May 5 as the default model for all ChatGPT users, replacing GPT-5.3 Instant. The headline stat: 52.5% fewer hallucinated claims on high-stakes prompts in medicine, law, and finance — with no latency penalty.
The release also adds enhanced personalization from past chats, connected files, and Gmail (rolling out to Plus and Pro first), plus new Memory sources across all consumer plans. The model is available in the API as chat-latest.
OpenAI released three new realtime voice models into the API: GPT-Realtime-2 for smarter live voice reasoning, GPT-Realtime-Translate for multilingual speech, and GPT-Realtime-Whisper for streaming transcription.
Separately, OpenAI is rolling out GPT-5.5-Cyber in limited preview to vetted cybersecurity teams — one month after Anthropic's Mythos debut, framing it as a defensive countermove to AI-enabled threats. The company is also making its most advanced models available to all vetted government levels for national security use.
cnbc.comWith Google I/O keynote set for May 19, the industry is watching for what's widely expected to be a Gemini 4 unveil — a major model update with native agentic capabilities, and possibly built-in image and video generation.
Google is also expected to announce Proactive Assistant, a feature that surfaces personalized suggestions from connected Google services without prompting. Android 17 will also feature prominently, with deeper Gemini integration baked into the OS. The Android Show pre-keynote drops on May 12 — tomorrow.
io.googlexAI's Grok 4.1 Fast is now available in Microsoft Copilot Studio for US-based makers — once enabled by org admins. The model is designed for large-context reasoning, deep tool use, and complex multi-step workflows, expanding the options available to enterprises building custom Copilot agents.
The integration follows xAI's earlier Grok 3 debut on Azure and GitHub Copilot's Grok Code Fast 1 preview, cementing a multi-vendor AI palette strategy at Microsoft. xAI simultaneously launched Grok Business and Enterprise tiers, with full data privacy guarantees — xAI won't train on organizational inputs.
microsoft.comNIST's Center for AI Standards and Innovation announced agreements with Google DeepMind, Microsoft, and xAI giving the US government the ability to evaluate AI models before public release — building on earlier partnerships with OpenAI and Anthropic from 2024.
The move signals a maturing US AI governance posture, prioritizing early access and testing over post-hoc regulation. The program aims to detect AI-enabled cybersecurity threats before they reach the wild — especially relevant given the emergence of restricted frontier models like Mythos and GPT-5.5-Cyber.
cnbc.comServiceNow is offering its AI Control Tower governance layer free for one year — a stated $2 million value — to any enterprise ready to deploy agentic AI. The pitch: before an autonomous agent deletes something critical, you need a kill switch.
The backdrop is stark: nearly all executives (97%) say they've deployed AI agents in the past year, yet only 1 in 10 has built anything truly autonomous. As agents gain permissions, call APIs, and write to databases, governance tooling is fast becoming a prerequisite rather than an afterthought. ServiceNow and Accenture also launched a joint Forward Deployed Engineering program to help enterprises move from pilot to production at scale.
fortune.comRather than converging, Claude Code, Cursor, and OpenAI Codex have staked out three distinct paradigms: terminal-first agents, IDE-native editors, and async cloud workers — respectively. Most production teams in 2026 use two or three together.
Cursor 3 (launched April 2) added a dedicated Agents Window that lets developers run multiple agents across repos simultaneously. OpenAI's Codex moved to token-based pricing in April. Claude Code remains the choice for deep codebase context and complex refactors. The tools are composable rather than competitive.
thenewstack.io