Real-time AI intelligence

Superpower Daily

The AI economy, before it hits the timeline.

OpenAI Proposes November 12 End to Cursor Model Access After SpaceX Deal

Cursor’s appeal is a single coding interface for several model providers. OpenAI’s proposal would remove one of those direct connections, with the migration path still undefined.

Open source

Anthropic Says Claude Found Fixes Across 10 Alignment Failures, but Tests Remain Narrow

10 hours ago

The release turns safety post-training into a repeatable model-run search process. Its value now depends on whether those benchmark gains survive broader tests and later training.

Security risk

TeamPCP’s Trivy Attack Hit 2,500 Organizations; AI Tool Backdoors Are Claimed

11 hours ago

The LiteLLM-linked exposure shows how a poisoned dependency can yield reusable developer access. Claims of persistence in AI coding-tool settings would make cleanup harder, but they remain uncorroborated.

Launch

Gnani Launches 30B Evon Model and Self-Hosted Agent Stack for Indian Institutions

19 minutes ago

The open-weight release couples Indian-language support with tools intended to keep sensitive data inside a customer’s own infrastructure.

Platform shift

OpenAI’s Windows Client Contains Gated Office Task Threads for ChatGPT

8 hours ago

Benchmark

Meta’s EvoHarness-RL Takes Qwen3-8B to 96.9% on ALFWorld, Near Claude Opus 4.5

11 hours ago

News wire

Latest updates

Updated hourly

See all latest

Benchmark

Lemmalog Turns Agent Memory Into Datalog State, Cutting LongMemEval Context 38-Fold

3 hours ago

The experimental system moves long-running agent work away from recalling old notes and toward maintaining conclusions with explicit support. Its early benchmark results suggest a substantial context-saving path, while leaving the probabilistic extraction step as the central weakness.

02Open sourceLightwheel Releases 10,000 Hours of Human Video for Robots, With 90,000 Still to Come5 hours ago

03BenchmarkPerplexity Search API Sweeps Artificial Analysis Test, but Medium Context Wins7 hours ago

04Platform shiftSageMaker Feature Store Adds 25-Record Batch Writes and a Record-Discovery API8 hours ago

Coverage desk

Models

Capability, pricing, benchmarks, and model strategy.

All Models

Launch

Tencent Releases Open-Source Hy4, a 770B Model That Activates 49B Parameters Per Request

16 hours ago

Tencent is pairing a large mixture-of-experts model with plans for product integration, while warning that the preview can spend too long working through difficult questions.

Launch

Amap Releases ABot-Recon, a 12-Frame 3D Mapper for 10,000-Frame Video Runs

21 hours ago

The model’s fixed local context could reduce the memory burden of long reconstruction jobs, but its headline speed is a data-center benchmark and its weights are not cleared for commercial deployment.

Benchmark

Netflix’s GenRec Pairs a 0.006% Gain With Catalog-Bound LLM Ranking

22 hours ago

The online result is statistically significant but too opaque to measure as a product outcome. The clearer contribution is a recommendation architecture that constrains an LLM to available titles while managing inference cost.

Benchmark

ReViSQL-K2.6 Hits 91.37% on Cleaned SQL Test, but Not a Live Database

23 hours ago

The result argues that a focused model, repaired examples, and domain-specific rewards can rival elaborate SQL-agent stacks. Its remaining test is whether that recipe transfers from curated academic data to corporate databases.

Superpower Daily

OpenAI Proposes November 12 End to Cursor Model Access After SpaceX Deal

Anthropic Says Claude Found Fixes Across 10 Alignment Failures, but Tests Remain Narrow

TeamPCP’s Trivy Attack Hit 2,500 Organizations; AI Tool Backdoors Are Claimed

Gnani Launches 30B Evon Model and Self-Hosted Agent Stack for Indian Institutions

OpenAI’s Windows Client Contains Gated Office Task Threads for ChatGPT

Meta’s EvoHarness-RL Takes Qwen3-8B to 96.9% on ALFWorld, Near Claude Opus 4.5

Latest updates

Lemmalog Turns Agent Memory Into Datalog State, Cutting LongMemEval Context 38-Fold

One clear read on the AI moves worth knowing.

Models

Tencent Releases Open-Source Hy4, a 770B Model That Activates 49B Parameters Per Request

Amap Releases ABot-Recon, a 12-Frame 3D Mapper for 10,000-Frame Video Runs

Netflix’s GenRec Pairs a 0.006% Gain With Catalog-Bound LLM Ranking

ReViSQL-K2.6 Hits 91.37% on Cleaned SQL Test, but Not a Live Database

Midjourney Opens V8.2 Image-Editing Tests With Up to Four References

MIT Builds PottsMPNN to Model Protein Stability Beyond Native-Sequence Matching

Cohere’s Parse 5 Turns Enterprise Documents Into Markdown for $1.50 per 1,000 Pages

Gemini Omni 1.1 Flash Adds 40-Second Scene Extensions and 4K Finishing Controls

Model-Serving Platforms Reach 6.1% of AI-Spending Businesses, While Frontier Spend Holds

Products

ICE Plans $1M–$2M SPOT Robot Buy for Hazard Assessments and Remote Inspection

Visa’s VVAH Adds AI Patch Writing Across 11 Stages, Leaving Merge Decisions to Humans

Oracle Adds AI Studio Skill to Fusion 26C, Deploying Pro-Code Agents Inside Fusion Apps

OpenAI’s Codex Client Prepares App Allowances, but No Tested Pro Entitlement Is Live

Harvard Uses 7 AI Faculty Avatars in a $699 Founder Bootcamp, While Keeping Live Advice Human

ChatGPT Lets Temporary Chats Use Memory, Then Convert Them to Regular Chats

Deepgram Sends SageMaker Billing and GPU Metrics to CloudWatch Without Container Egress

Google AI Mode Adds Fare Alerts in 180+ Countries and U.S. Hotel Booking

Hugging Face Sells a $399 Microduck Robot With Its Training Stack on GitHub

OpenAI’s Astra is claimed to solve a math problem

OpenAI slows reinforcement learning after agent breach

Harvard’s AI Faculty, Agent Tools, and the Limits of Delegation

The Signal: Cheaper AI Workers, Cost-Conscious Buyers, and Control Gaps

OpenAI Pauses Frontier Training as Permissioned AI Takes Center Stage

Stripe’s OpenRouter Deal Puts AI Routing Beside the Billing Stack

The Week AI Hit Its Operational Boundaries

Tools

NVIDIA’s TensorRT Model Connect Skips ONNX, but x86_64 Users Must Build From Source

Salesforce Uses SageMaker Placement Controls to Meet 2-AZ Agentforce Standard, Keep 8x Savings

Intel Releases 20 Agent Skills for Arc GPUs, With an Open-Port Security Catch

Perplexity’s GitHub-Slack Agent Uses One API Call and a One-Reply Guardrail

Databricks Says Async Save Cut 20B-Parameter PyTorch Checkpoints to 9 Seconds on 32 H100s

Amazon Quick Connects to fal Through MCP, Making Creative Approval Steps Reusable

Snowflake Open-Sources Semi-Persistence With Sub-Second Single-GPU vLLM Swaps

Experiential Open-Sources an Agent Router That Learns From Traffic, but Results Are Unverified

OpenAI’s 10-Day WebMCP Challenge Pushes Websites to Expose Tools for AI Agents

Startups

IndiGo Ventures Joins Sarvam’s $1.5B Series B, Linking Airline Work to Ownership

DeepSeek Reportedly Seeks $7.4 Billion at $74 Billion Valuation for Compute Buildout

Portica AI Files $450,000 SAFE Offering Fully Sold to Three Investors

Instinct Raises $250M at $2.5B for a Private-Beta Agent With Deep Access

Deep Cogito Raises $43M to Build AI Models Enterprises Can Own

Emerald AI Raises $150M at $1.05B to Make Data Centers Flexible Grid Loads

Capital moving through AI.

Funding

Acquisitions

You've seen the signals. Get the complete briefing tomorrow.

Business

Anthropic Explored $7B MatX Deal, but Keeps a Multi-Chip Route to Custom Silicon

Decathlon Deploys Chronos-2, Cuts 12-Week WAPE by 11–15 Points

Nvidia Says AI Data-Center Power Can Take 2–3 Years; Bloom Claims a 55-Day Route

Nasuni Buys DryvIQ, Adds Governance Across 40-Plus Data Repositories

Amazon Plans $220B Capex as AWS Revenue Grows 37%, Testing Returns

Salesforce Says Agentforce Passed $1.5B ARR as It Pushes Consumption Into Premium Plans

IREN Takes $639M Mining Writedown as AI Cloud Revenue Rises Nearly 8x

OpenAI Hires Meta’s Devanathan for Southeast Asia and Australia Growth, Policy

Meta Tests Data-Center Repair Robots, but Cable Bots Still Need Human Supervision

Policy

Coughlan and Lucas Join About 80 in UK Voice-Ownership Push Against AI Cloning

Anthropic-Backed Group Sends $500,000 to a Pro-Markey PAC as AI Defines Senate Primary

Meta’s Up-to-$18 Billion Deal Carves Out Under-13 Data for Age-Assurance Testing

Judge Vacates Pentagon’s Anthropic Blacklist and Orders Defense Department Retraction

100 Firms Warn AI Cyberattacks Will Grow Within Months, Putting Hospitals and Utilities at Risk

Pennsylvania Poll Finds 79% Reject Local Data Centers as 84% Back Tighter AI Rules

NSA Seeks All Commercial AI Models as White House Offers 30-Day Pre-Release Tests

Bill Gates Calls for Taxes on AI Tokens and Robots, With Jobs Reserved for Humans

US Data Centers Could Use 1.125 Trillion Liters of Water by 2030, Straining Dry Regions

Culture

JAMA Paper Says Autonomous AI Could Surpass Physicians and AI-Assisted Care by 2030