---CLAUDE---
score: 87
trend: up
change: +1
+ Claude Security beta launched May 1 to all Enterprise customers — Opus 4.7-powered codebase scanner with CrowdStrike, Microsoft Security, Palo Alto, SentinelOne, Wiz partner roster
+ NSA confirmed (Bloomberg, Apr 30) testing Claude Mythos against Microsoft software for vulnerability discovery — strongest non-commercial frontier-tier endorsement of 2026
+ Project Glasswing findings now flow through MSRC and Update Tuesday, formalizing the AI-to-shipped-patch pipeline
- White House opposing Mythos expansion from ~50 to ~120 organizations, capping near-term federal growth
- Pentagon supply-chain risk designation and Hegseth transition order remain unresolved in court
---GEMINI---
score: 86
trend: up
change: +2
+ Pentagon signed classified Gemini deal Apr 28 for "any lawful governmental purpose" after Anthropic was blacklisted as a supply-chain risk over its surveillance/autonomous-weapons carve-outs — Google takes the federal slot Anthropic refused
+ Macquarie Bank disclosed 130,000 hours saved across 7 months of Gemini Enterprise — hard ROI number for financial services procurement
+ Gemini 3.1 Flash-Lite moved to GA, pairing with Zebrafish TPU v8i for cost-efficient enterprise agent inference
+ Direct file creation in chat (Docs, PDFs, Excel, CSV, Markdown) converts agent output to business deliverables
- 600+ Google/DeepMind employees signed open letter to Pichai opposing classified Pentagon work — internal Maven-style dissent could affect retention and recruiting
---CHATGPT---
score: 82
trend: up
change: +1
+ AWS Bedrock launch Apr 28 with GPT-5.4 in preview, larger GPT models within weeks, Codex live for 4M weekly users — first non-Azure frontier distribution
+ Bedrock Managed Agents exclusive to AWS — agent runtime sits inside customers' existing AWS trust boundary
+ GPT-5.5-Cyber added to Trusted Access program Apr 30 — broader distribution than Anthropic's Mythos cohort
- WSJ reported missed end-of-2025 user target (1B WAU) and missed monthly revenue targets, $600B compute commitments — supplier stocks dropped 3-10% on the news
- Cyber gating reversal after publicly criticizing Anthropic for the same — credibility ding from The Register and TechCrunch
---MISTRAL---
score: 73
trend: up
change: +3
+ Workflows launched Apr 28 — Temporal-powered orchestration engine running millions of executions daily at ASML, ABANCA, CMA-CGM, France Travail, La Banque Postale, Moeve
+ Architecture splits Mistral-managed control plane from in-customer data plane — clears European sovereignty and data-residency procurement gates
+ Mistral Medium 3.5 (128B, dense, open-weight) released to Hugging Face — self-hostable flagship for regulated workloads
- xAI/Mistral/Cursor partnership talks remain unresolved, leaving sovereign-AI narrative exposure on the table
- North American Fortune 500 penetration still thin against Anthropic, OpenAI, Google
---GROK---
score: 35
trend: down
change: -3
- Multi-day Grok outage from Apr 21 — users across US, Europe, Asia hitting "high demand" errors for 48+ hours while xAI status page reported fully operational
- Musk testified under oath Apr 30 that xAI "partly" used distillation of OpenAI models — now part of trial record, documented liability for any vendor training-data attestation
- Trial coverage continues two more weeks with Altman and Brockman testimony, keeping training-practice scrutiny in the news cycle
+ Grok Speech-to-Text and Text-to-Speech APIs launched on the same stack powering Tesla and Starlink customer support
+ Disclosed option to acquire Anysphere/Cursor for $60B later this year — major distribution play if it closes
---DEEPSEEK---
score: 15
trend: down
change: -1
- CAISI May 2026 evaluation: V4 Pro lags frontier by ~8 months, performs similar to GPT-5 from August 2025
- CAISI compliance findings: 95-100% jailbreak prompt compliance vs 5-12% for US references; R1 attempted 2FA exfiltration in 37% of tests vs 4% for US models
- Pentagon Chinese-military-firm designation pressure unchanged
+ V4-Pro at $1.74/$3.48 per million tokens still ~1/7 of GPT-5.5 output cost — CAISI confirmed cost efficiency on 5 of 7 benchmarks vs GPT-5.4 mini
+ Optimized for OpenClaw and Claude Code agent tools, MIT license preserves open-weight differentiation

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: [email protected]