May 8 AI News: Anthropic Doubles Claude Code Limits, ElevenLabs Crosses $500M, OpenAI Ships Three Voice Models

Anthropic Doubles Claude Code Rate Limits via SpaceX Colossus Deal

Anthropic doubled Claude Code's rate limits overnight by partnering with SpaceX's Colossus One data center, adding over 300 megawatts of new capacity — the equivalent of more than 220,000 Nvidia GPUs — in under a month. The move comes after developers had reported hitting output limits during peak hours. The Colossus facility, originally built for xAI's Grok training workloads, is now providing burst compute for Anthropic's developer-facing products. No pricing changes were announced alongside the rate limit increase.

ElevenLabs Crosses $500M ARR, Cuts API Pricing After Series D

ElevenLabs reported surpassing $500 million in annual recurring revenue following a Series D funding round, and simultaneously announced significant price reductions across its voice and agentic API services. The cuts target developers building voice agents at scale — a segment ElevenLabs has been aggressively pursuing against competitors including OpenAI's Realtime API and PlayAI. No specific percentage cuts were disclosed in the announcement, but the company confirmed the reductions apply to both per-character and per-minute billing tiers.

OpenAI Ships Three Voice Models in the API

OpenAI launched three new voice-focused models via the API: GPT-Realtime-2 (bringing GPT-5-class reasoning to real-time voice), GPT-Realtime-Whisper (optimized for transcription workloads), and GPT-Realtime-Translate (speech-to-speech translation). The releases expand OpenAI's position in the voice API market, where it competes directly with ElevenLabs, Deepgram, and AssemblyAI. GPT-Realtime-2 is the most significant of the three — it applies the reasoning capabilities of the GPT-5 family to low-latency voice, which prior Realtime API models lacked.

GPT-5.5-Cyber Released to Vetted Security Teams

OpenAI is rolling out GPT-5.5-Cyber, a security-focused variant of GPT-5.5, in limited preview to vetted cybersecurity teams. The model is positioned as a direct response to Anthropic's Claude Mythos Preview, which has been used under Project Glasswing to identify zero-day vulnerabilities across major operating systems and browsers. OpenAI has not confirmed a public release timeline for GPT-5.5-Cyber. Access is currently gated through a separate application process for qualifying security organizations.

White House Drafts Executive Order to Pre-Vet AI Models

National Economic Council Director Kevin Hassett confirmed on May 7 that the White House is drafting an executive order requiring AI models to be vetted before public release — drawing a direct comparison to the FDA drug approval process. The order is described as a direct response to the dual-use risks raised by Anthropic's Mythos model, which can autonomously identify network vulnerabilities. The Commerce Department has already expanded its voluntary pre-release testing program to include Google, Microsoft, xAI, OpenAI, and Anthropic. The six research teams who disclosed security exploits against OpenAI Codex, Claude Code, GitHub Copilot, and Google Vertex AI between 2025 and 2026 added further urgency to the policy push — every exploit targeted credentials, tokens, or permissions rather than the underlying models themselves.

Pennsylvania Sues Character.AI Over Medical Impersonation

Pennsylvania filed suit against Character.AI, alleging the platform misrepresented AI personas as qualified medical professionals to users who were seeking health guidance. The case is among the first state-level actions to specifically target AI persona design rather than content moderation failures. It follows a pattern of state attorneys general moving ahead of federal AI legislation, with Connecticut's comprehensive AI bill and Iowa's chatbot safety law both advancing in the same week.

Also on May 8

xAI launched the Grok Imagine Quality Mode API for photorealistic image generation and integrated Connectors for Google Workspace and Outlook into the Grok web interface. US-China AI talks are gaining momentum ahead of a planned Trump-Xi summit, with Treasury Secretary Scott Bessent leading the American side on AI policy discussions. The Fivetran 2026 Agentic AI Readiness Index found that only 15% of organizations have a data foundation capable of safely running AI agents at production scale, despite nearly 60% already investing millions in the technology.

For full May 2026 AI coverage see our AI News May 2026 hub. Next: May 9 AI News.