News

Latest: November 3, 2026

1. NVIDIA/auto-ontology

   github.com

   Open-sources an Apache-2.0 Python framework from NVIDIA that builds semantic knowledge graphs from enterprise metadata to reconcile cross-system metrics, validate queries with lineage provenance, and expose context via MCP, CLI, and API.

2. GDPx: Measuring Frontier Model Capabilities Across the Most Valuable Domains

   newsletter.semianalysis.com

   Benchmarks frontier models across 11 high-ROI domains—including chip design, cyber, bio, quant trading, RSI, alignment, and software engineering—where the top model scores 12% and OpenAI and Anthropic lead despite safety classifiers sandbagging Opus 5.5 and GPT-6 Astra.

3. Lessons From Red-teaming Auto-review, OpenAI's Blocking Monitor

   lesswrong.com

   Uncovers 11 exploit vectors that bypass OpenAI Codex's sandbox and synchronous Auto-review monitor via coverage gaps, prompt injections, truncated evidence, and stale-approval timing attacks, prompting OpenAI fixes and controls against trusting broad code loaders like pytest.

4. The AI cost of JavaScript frameworks

   nolanlawson.com

   Benchmarks Claude Code with Opus 5.5 across 12 JavaScript frameworks over five runs using official MCP servers and skills, finding median build costs from $2.67 (Hono+htmx), $2.76 (Astro), and $2.82 (Nuxt) up to $4.28 (SvelteKit), $4.65 (Gracile+Lit), and $4.97 (Angular).

5. Postgres 19 is late, and it's AI's fault

   rhaas.blogspot.com

   Explains how LLM-assisted bug reports drove a 60% jump in back-patched commits to REL_19_STABLE (1,169 vs. 724) and 74% to REL_18_STABLE (433 vs. 248), delaying PostgreSQL 19 past October and forcing late reverts of FOR PORTION OF, SQL/PGQ, and partition merge/split features.

6. ARTEX and Friends: Enter Agentic Offensive Frameworks

   cloud.google.com

   Details Mandiant and Google Threat Intelligence Group incident response cases where threat actors weaponized open-source multi-agent pentesting frameworks including ARTEX, Hexstrike, and CAIRN to automate reconnaissance, exploitation, and attack-path chaining at machine speed.

7. The Rise of LLMJacking: How Threat Actors Monetize Compromised Cloud and Model Resources

   cloud.google.com

   Analyzes 2026 Mandiant incident response engagements where threat actors hijacked enterprise cloud LLM credentials across finance, retail, tech, and transport, generating fraudulent API spikes exceeding $100,000 per day and up to $1 million per victim alongside an underground credential market.

8. North Korea Steals USD 6.36 Billion From Binance in the Largest Hack in History

   trmlabs.com

   Attributes the October 29, 2026 theft of $6.36 billion from Binance across 10 blockchains to North Korea's TraderTraitor via compromised hot wallet signing infrastructure, with $2.7 billion frozen while the attacker launders roughly $2.4 billion through market makers and lending protocols.

9. Binance Hack: What Can a Compromised Laptop Reach?

   blog.gitguardian.com

   Examines how a compromised DevOps engineer laptop initiated the $6.3 billion Binance breach via an internal service container without leaking wallet private keys, highlighting lateral movement risks from CLI token caches, SSH keys, config credentials, and secrets left by AI coding agents.