COGNOSCERE Daily Tech Review — Issue T123 · Friday, May 22, 2026

Or visit Intelligence Overview for deeper analysis.

ACTGitHub Confirms Breach of 3,800 Internal Repos via Malicious Nx Console VS Code Extension

GitHub confirmed that threat actor TeamPCP exfiltrated data from approximately 3,800 internal repositories after a GitHub employee installed a backdoored version of the Nx Console VS Code extension, which was live on the marketplace for only 18 minutes. The breach exposed internal source code for products including Actions, Copilot, and Dependabot; GitHub rotated credentials and stated no customer repositories were confirmed impacted, though the investigation remains ongoing.

BleepingComputer · Cybersecurity Incidents · Relevance: 1.0 · Source →

supply-chain attack, developer tooling, VS Code, credential theft, source code breach, TeamPCP, software supply chain

ACTCISA Contractor Leaked Highly Sensitive AWS GovCloud Keys on GitHub

A contractor for CISA (operated by Nightwing) maintained a public GitHub repository called 'Private-CISA' from November 2025 through mid-May 2026, inadvertently exposing 844 MB of data including administrative credentials for three AWS GovCloud accounts, plaintext passwords for dozens of internal CISA systems, and access to CISA's internal software build pipeline (Artifactory). GitGuardian researcher Guillaume Valadon, who discovered the leak, called it the worst he had witnessed in his career; CISA confirmed no sensitive data was confirmed compromised but is investigating.

Krebs on Security · Cybersecurity Incidents · Relevance: 0.9 · Source →

government security, secrets management, AWS GovCloud, credential exposure, supply chain, CISA, cloud security

PREPAREAnthropic to Pay SpaceX Nearly $45 Billion for Computing Resources Through May 2029

Disclosed in SpaceX's IPO S-1 filing, Anthropic has committed to paying SpaceX $1.25 billion per month through May 2029 — totaling nearly $45 billion — for access to compute capacity across SpaceX's Colossus 1 and Colossus 2 AI data centers in Memphis, which together provide over 220,000 NVIDIA GPUs. The deal, which either party can terminate with 90 days' notice, is the largest publicly disclosed AI compute contract to date and positions SpaceX as a significant AI infrastructure provider alongside hyperscalers.

Bloomberg · Data Infrastructure · Relevance: 0.9 · Source →

AI infrastructure, compute, cloud services, GPU, enterprise AI, SpaceX IPO, capacity planning, AI economics

PREPAREMicrosoft Open-Sources RAMPART and Clarity AI Agent Security Testing Frameworks

Microsoft released two open-source tools for hardening AI agent security during development: RAMPART, a Pytest-native red-teaming framework covering adversarial test cases for cross-prompt injection, data exfiltration, and other harm categories, and Clarity, a structured design-review tool that pressure-tests system assumptions before code is written. The tools extend Microsoft's existing PyRIT framework and are intended to shift AI security testing earlier in the agent development lifecycle.

TLDR InfoSec · Cybersecurity Policy · Relevance: 0.8 · Source →

AI security, red teaming, agentic AI, open source, prompt injection, developer tools, AI governance, cybersecurity

WATCHCloudflare and Anthropic Integrate Claude Managed Agents with Cloudflare Infrastructure

Cloudflare and Anthropic announced the integration of Claude Managed Agents with Cloudflare's edge infrastructure, enabling developers to deploy AI agents with enhanced security features including customizable proxies, private service connectivity, and lightweight V8 isolates instead of full microVMs for faster and cheaper scaling. The integration includes out-of-the-box capabilities such as browser control with session recording, email functions per agent, and native connections to Cloudflare Workers AI and R2 storage.

TLDR DevOps · Cloud Computing · Relevance: 0.8 · Source →

cloud computing, agentic AI, edge computing, AI infrastructure, enterprise software, developer tools, security

PREPAREGoogle Launches Gemini 3.5 Flash at I/O 2026: Agentic AI at Enterprise Scale

Google released Gemini 3.5 Flash at I/O 2026, describing it as its strongest agentic and coding model to date, running 4x faster than comparable frontier models at roughly half the cost and up to 12x faster within its Antigravity 2.0 agentic development platform. The model is now generally available across the Gemini app, AI Mode in Search, Gemini Enterprise, and the Gemini API, while Google also introduced Gemini Spark (a 24/7 personal AI agent), Gemini Omni (multimodal video generation), and agentic capabilities in Search; Google expects enterprises running one trillion tokens per day could save over $1 billion annually by shifting 80% of workloads to Flash.

TechCrunch · AI Model Releases · Relevance: 1.0 · Source →

LLM, agentic AI, enterprise AI, Google I/O, multimodal, model release, AI cost, developer platform

WATCHOpenAI Reasoning Model Autonomously Disproves 80-Year-Old Erdős Geometry Conjecture

OpenAI announced that an internal general-purpose reasoning model autonomously disproved Erdős's 1946 planar unit distance conjecture using techniques from algebraic number theory, representing the first time an AI system has autonomously resolved a prominent open problem in mathematics. The proof was independently verified by external mathematicians including Fields medalist Tim Gowers, Noga Alon, and Thomas Bloom, who endorsed the result as publishable in the Annals of Mathematics.

OpenAI · AI/ML Tooling and Frameworks · Relevance: 0.8 · Source →

AI reasoning, mathematical discovery, general-purpose AI, frontier capability, AI safety, scientific AI, Level 4 AI

PREPARESAP Launches Business AI Platform and API Policy at Sapphire 2026, Positioning Itself as Enterprise AI Gatekeeper

At SAP Sapphire 2026, SAP unveiled its Business AI Platform — consolidating SAP BTP, SAP Business Data Cloud, and Business AI — along with over 50 Joule AI agents for autonomous ERP workflows and an AI Agent Hub for centralized governance. Concurrently, SAP's new API Policy (enforcement begins June 9, 2026) restricts third-party AI agents from accessing SAP data outside SAP-controlled pathways, with Forrester warning CIOs to scrutinize the concentration risk and undisclosed 2027 pricing implications.

Forrester · Business Intelligence Platforms · Relevance: 0.9 · Source →

enterprise AI, ERP, agentic AI, AI governance, vendor lock-in, data access, decision support, autonomous enterprise, SAP

WATCHGoogle Co-Scientist Published in Nature: Hypothesis Generation Tool Now Available to Researchers

Google published its AI Co-Scientist research in Nature, debuting a Hypothesis Generation feature that uses Gemini-powered agents competing in 'idea tournaments' to surface novel hypotheses for biology labs, with one lead reportedly cutting a liver-fibrosis-related scarring signal by 91% in a Stanford study. Google simultaneously launched Gemini for Science, pairing Co-Scientist with AlphaEvolve and NotebookLM, with waitlist access now open for individual researchers.

Nature · Decision Frameworks · Relevance: 0.8 · Source →

AI for science, decision support, hypothesis generation, scientific AI, biotech, agentic AI, enterprise AI