Thursday, May 7, 2026

Anthropic Enterprise · Finance

Anthropic Launches Finance Agents and Claude Opus 4.7 for Wall Street

Anthropic unveiled a suite of 10 pre-built AI agents targeting the most labor-intensive banking workflows — pitchbooks, earnings analysis, credit memos, KYC, underwriting, and insurance claims — at an invite-only briefing in New York. The agents run natively inside Claude for Microsoft 365, operating directly within Excel, PowerPoint, and Word.

Powering the suite is the new Claude Opus 4.7, which leads Vals AI's Finance Agent benchmark at 64.4% and tops GDPval-AA for economically valuable knowledge work. A new Moody's integration supplies proprietary credit ratings and data on over 600 million companies. Anthropic said Claude is already in production at JPMorgan Chase, Goldman Sachs, Citi, AIG, and Visa.

64.4%

Vals Finance Benchmark

600M+

Moody's companies covered

10

Pre-built agent templates

Fortune coverage

Google Microsoft xAI Policy · Security

NIST's CAISI Expands Pre-Deployment AI Testing to Google, Microsoft, and xAI

The Commerce Department's Center for AI Standards and Innovation (CAISI) signed frontier AI testing agreements with Google DeepMind, Microsoft, and xAI. The three companies join OpenAI and Anthropic in submitting models — including versions with reduced or removed safeguards — for pre-release evaluation in classified environments.

Evaluators from across the US government participate via the CAISI-convened TRAINS Taskforce, an interagency body focused on AI national-security concerns. CAISI said it has completed more than 40 evaluations to date, including assessments of unreleased models. The agreements were renegotiated to align with the Trump administration's AI Action Plan.

NIST announcement

Anthropic Security · Policy

Anthropic's Mythos: The Cybersecurity Model Too Dangerous to Release — and Already Breached

Claude Mythos Preview, announced in early April, is capable of discovering zero-day vulnerabilities across major operating systems and chaining software bugs into multi-step exploits — a feat previously achievable only by top human hackers. Anthropic restricted access to roughly 40 approved tech firms, excluding most central banks and governments.

The model was breached on its very launch day: an unauthorized group gained access through a third-party vendor environment, facilitated by a contractor with shared API keys. A separate diplomatic row has emerged as Anthropic negotiates with EU officials over access, while operational technology providers express frustration at their exclusion. OpenAI subsequently restricted its own "Cyber" model after initially criticizing Anthropic's approach.

Zero-day discovery Multi-step exploit chaining Project Glasswing EU access negotiations

TechCrunch report

Industry IPO · AI Chips

Cerebras Targets $26.6B Valuation in 2026's Largest Tech IPO — Backed by OpenAI's Ecosystem

AI chipmaker Cerebras Systems plans to sell 28 million shares at $115–$125 each, targeting a raise of $3.5 billion and a market cap of $26.6B at the high end. The Nasdaq listing under ticker CBRS is set for May 13. Early demand is reported at $10B against the $3.5B offer.

Cerebras's Wafer-Scale Engine 3 claims faster inference and lower power than GPU alternatives. The company has deep OpenAI ties: a $20B+ compute deal through 2028, a $1B loan from OpenAI secured by warrants, and angel investors including Sam Altman, Greg Brockman, and Ilya Sutskever.

$26.6B

Target valuation

$3.5B

Raise target

$20B+

OpenAI compute deal

TechCrunch coverage

OpenAI Business · IPO

OpenAI Eyes Q4 2026 IPO at $852B — But CFO Urges Caution Amid $14B Projected Losses

OpenAI has surpassed $25B in annualized revenue and is reportedly laying groundwork for a public listing in Q4 2026, targeting an $852B valuation and a raise of over $60B. The company plans to allocate IPO shares to retail investors.

However, CFO Sarah Friar has suggested a possible delay to 2027, as the company remains heavily loss-making — internal projections point to $14B in losses in 2026 alone — with profitability not expected until around 2030. OpenAI also dropped plans to spin off its robotics and hardware divisions, choosing instead to focus on core revenue-generating products.

$25B+

Annualized revenue

$852B

IPO valuation target

$14B

Projected 2026 losses

Gizmodo report

Meta Model Release · Strategy

Meta Breaks from Open-Source Playbook with Proprietary "Muse Spark" LLM

Meta unveiled Muse Spark, its first flagship proprietary large language model built under Chief AI Officer Alexandr Wang's newly formed Superintelligence Labs. The model — a deliberate departure from Meta's open-source Llama strategy — delivers competitive performance on multimodal perception, reasoning, health, and agentic tasks at a fraction of the compute cost of Llama 4 mid-size variants.

The announcement came alongside Meta's disclosure of $115–135B in AI capital expenditure for 2026, nearly double last year's spending. Analysts noted that Meta and Microsoft were the only major AI companies whose capex announcements failed to convince investors it was paying off, while Google's disclosure drove a share price rally.

Fortune coverage

Microsoft Google Enterprise · Agents

Microsoft and Google Push AI Agent Governance into Enterprise IT Mainstream

Microsoft's Agent 365 — generally available for commercial customers as of May 1 — lets IT teams discover, govern, and secure AI agents across Microsoft, third-party SaaS, cloud, and on-prem environments. A new Intelligent Purview service extends data-loss prevention to AI prompts and responses in real time, blocking agents from leaking credit-card numbers or IP-sensitive material in outputs.

Google separately launched a centralized AI control center for Workspace giving administrators unified visibility over AI usage, security settings, data protection controls, and privacy safeguards. Google Cloud also unveiled the Gemini Enterprise Agent Platform, billed as a unified OS for building, scaling, governing, and optimizing AI agents across the enterprise.

Agent 365 GA Intelligent Purview DLP Gemini Enterprise Agent Platform Workspace AI Admin

Computerworld

Cursor OpenAI Anthropic Developer Tools · Benchmarks

AI Coding Tool Wars: Codex Hits 3M Weekly Users as Cursor 3 Challenges Claude Code

The AI coding landscape is consolidating around three major stacks. OpenAI Codex has surpassed 3 million weekly active users, up from 2M just a month earlier. Cursor 3, an agent-first redesign, launches on the Kimi K2.5 base, scoring 73.7 on SWE-bench Multilingual and competing directly with Claude Code and Codex for developer mindshare.

Despite the competition, Claude Code leads developer surveys — a Pragmatic Engineer poll of 906 engineers gave it a 46% "most loved" rating. Its structural edge is context: up to ~1M tokens on Max/enterprise plans vs. Cursor's practical 70–120k. Cursor wins on interactive speed with sub-second Tab prediction. All three tools now support Claude Opus 4.7 at identical underlying pricing.

3M+

Codex weekly users

46%

Claude Code "most loved"

73.7

Cursor 3 SWE-bench ML

The New Stack