Anthropic
🕐 2 min read
Meta and Anthropic in Talks for $10 Billion Compute Lease Deal
Meta and Anthropic are negotiating a two-year deal where Meta would lease excess data center capacity to Anthropic for approximately $10 billion. The arrangement would make Meta simultaneously a competitor in the AI model market and a critical infrastructure supplier to one of its rivals.
⚡ Why It Matters: This deal reveals the structural reality of AI infrastructure: even well-funded labs cannot self-supply the compute needed for frontier models. The ability to access infrastructure is becoming as important as the ability to build models.
📰 New York Times →
Research
🕐 2 min read
Moonshot AI Releases Kimi K3: 2.8 Trillion Parameter Open-Weight Model
Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight model that achieved a 76% win rate against GPT-5.6 and Fable 5 on frontend coding benchmarks. Full public weights release is scheduled for July 27, making it available for anyone to download and fine-tune.
⚡ Why It Matters: Open-weight models are now viable competitors to closed-API models for organizations that value data privacy and customization. July 27 sets a new capability bar for what open-weight means.
📰 TechCrunch →
Research
🕐 2 min read
Thinking Machines Lab Launches Inkling: Mira Murati's First Open-Weight Model
Thinking Machines Lab launched Inkling, a 975 billion parameter open-weight multimodal model trained simultaneously on text, images, audio, and video. Designed for fine-tuning through the Tinker platform, it brings enterprise-grade multimodal capability to organizations without API dependency.
⚡ Why It Matters: Inkling represents a direct response to enterprise demand for controllable, customizable AI. Organizations can now fine-tune frontier-class multimodal models on proprietary data without sending that data to external APIs.
📰 Bloomberg →
Microsoft
🕐 2 min read
Microsoft Trains Salespeople to Pitch Against OpenAI and Anthropic
Bloomberg confirmed Microsoft has been running internal training sessions equipping its enterprise sales force with specific talking points to displace OpenAI and Anthropic. VP Jacob Andreou explicitly called Anthropic's Claude 'slower and less accurate' than Copilot in an internal meeting.
⚡ Why It Matters: The enterprise AI alliance era is over. Organizations with both Copilot and standalone AI subscriptions should audit usage and prepare for vendor-neutral evaluations rather than relying on vendor-provided comparisons.
📰 Bloomberg →
Governance
🕐 2 min read
DeepMind CEO Hassabis Calls for FINRA-Like AI Standards Body
Demis Hassabis publicly called for the US to establish an independent 'Frontier AI Standards Body' modeled on FINRA, with authority to set safety standards and coordinate industry-wide slowdowns. He aims for the body to be operational by end of 2026.
⚡ Why It Matters: Industry-led governance with mandatory safety standards is now a position held by one of the most credible voices in AI. The institutional design conversation is active, and formal structures may emerge within 12-24 months.
📰 Axios →
Industry
🕐 1 min read
TSMC Reports Record Q2 2026: $40.2B Revenue, 77% Net Income Growth
TSMC — the factory behind virtually every significant AI chip — reported all-time record results with $40.2 billion revenue, 77% net income growth, and 60.3% operating margins. The results confirm AI infrastructure demand is structurally real and outpacing supply.
⚡ Why It Matters: TSMC's margins confirm AI hardware demand is outstripping supply faster than capacity can expand. Every compute story this week — from Meta-Anthropic to model training costs — is downstream of this supply chain reality.
📰 TSMC Press Release →
OpenAI
🕐 2 min read
GPT-Red: OpenAI's AI Red-Teamer Cuts Jailbreak Success to Under 23%
OpenAI unveiled GPT-Red, an internal AI system that autonomously finds security weaknesses in other AI models. Applied to GPT-5.6, it reduced successful adversarial attack rates from over 90% to under 23% — the first confirmed case of AI auditing AI at production scale.
⚡ Why It Matters: GPT-Red raises the bar for what 'thoroughly tested' means. Organizations deploying AI should ask vendors what adversarial testing they perform — AI-powered red-teaming is becoming the standard for production-grade safety.
📰 MIT Technology Review →
Microsoft
🕐 1 min read
Microsoft Project Perception: AI Cybersecurity Tool Targets Anthropic's Mythos
Microsoft announced Project Perception, an AI-powered vulnerability detection platform using models from Microsoft, OpenAI, and Anthropic. Explicitly positioned as a cost-effective competitor to Anthropic's Mythos, it integrates with Microsoft Defender and ships by end of July.
⚡ Why It Matters: AI security is moving from specialty service to mainstream product category. Security professionals who develop AI-specific skills are positioning themselves in a rapidly growing specialization.
📰 TechRepublic →
Anthropic
🕐 1 min read
Claude Corps Application Window Closes — First Cohort Begins October 2026
The application window for Anthropic's Claude Corps fellowship closed July 17. The program offers 100 positions at $85,000 per year to work on mission-critical AI deployment in public sector, healthcare, education, and civil society — with no university degree required.
⚡ Why It Matters: Claude Corps signals that AI deployment specialists in mission-critical sectors are now valued enough for Anthropic to pay $85K with no degree requirement. Next cohort applications open early 2027.
📰 Anthropic →
Governance
🕐 1 min read
India MeitY Advisory Dispute: Reports of OpenAI/Anthropic Pause, Then Government Denial
ThePrint reported India's MeitY directed ministries to pause using OpenAI and Anthropic for cybersecurity functions, citing data sovereignty concerns. The government issued a formal denial within 24 hours, but the episode illustrates emerging AI sovereignty tensions.
⚡ Why It Matters: AI sovereignty policies are fragmenting the international AI market. Organizations providing AI services to government customers should map sovereignty policies by country as a distinct regulatory risk.
📰 ThePrint →