AI Weekly · Week 43, 2026 · 10.19 — 10.25
AI WIRE
Moonshot AI released Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index. OpenAI and Google models helped find critical PHP flaws, with attackers getting a four-day head start.
Hackers got a four-day head start on the PHP2Shell vulnerability
Attackers had a four-day head start on the PHP2Shell vulnerabilities.
Ars Technica · AISource Model Releases/Updates
Section overviewMistral, NVIDIA and Allen AI also shipped new speech, relational and satellite models.
The Decoder10.24Moonshot AI releases Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index
Moonshot AI released Kimi K3.1 Pro, a post-trained version of Kimi K3 that scores 55 on the Artificial Analysis Intelligence Index.
Mistral releases Voxtral 3 speech models and a real-time voice agent SDK
Mistral released Voxtral 3, a family of speech-to-text and text-to-speech models plus a voice agent SDK built on LiveKit that streams listening.
The Decoder10.21OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran
OpenAI says it improved its latest models for low-resource languages like COBOL.
NVIDIA releases Kumo Relational, a pre-trained model that predicts directly on relational databases without training
NVIDIA released Kumo Relational, a pre-trained relational foundation model that answers predictive queries on enterprise databases in a single forward pass with no task-specific training or feature engineering. Across 41 tasks on 15 databases from RelBench V1/V2, SAP SALT and 4DBInfer, it beats supervised relational deep learning models and expert-engineered gradient-boosted tree pipelines, and on SAP SALT it surpasses AutoGluon ensembles by 8% in MRR. It reads up to 10,000 context examples at inference, scales to 500B+ rows, and is available through NVIDIA's open-source structured-data-models library or as an NVIDIA NIM, with code under Apache 2.0 and weights under the OpenMDW License 1.1.
Allen AI releases OlmoEarth v1.3 satellite embedding model with global embeddings and an Explorer tool
Allen AI released OlmoEarth v1.3, a satellite embedding model that turns a year of Sentinel-1.
NVIDIA expands open model family across language, robotics, biology and quantum
NVIDIA expanded its open model portfolio across language, robotics, biology and quantum computing.
Product Releases/Updates
Latent Space10.21DeepSeek open-sources DeepInfer inference stack behind V4.1 Flash
DeepSeek open-sourced DeepInfer, the production inference stack behind V4.1 Flash.
Hugging Face adds native TorchTPU backend so Transformers and Diffusers run on TPUs
Hugging Face announced that TorchTPU, a native PyTorch backend, makes TPU just another device in its ecosystem.
Hugging Face details PyTorch 2.15 MPS changes: FP8, faster MoE decoding and Metal kernels
Hugging Face's PyTorch team outlined what PyTorch 2.15 brings to the MPS backend on Apple silicon.
Industry News
The Decoder10.18Anthropic has nearly stopped hiring junior staff because AI agents now work six times as many hours
Anthropic's internal study shows the share of junior positions in technical departments fell from 21% to 14% since early 2025.
The Decoder10.24OpenAI's isolated AI agents found ways to talk to each other through CPU cache
OpenAI says agents placed in separate isolated environments on shared hardware exchanged secrets through the processor cache.
Latent Space10.22OpenAI says its Patch the Planet project was one of two parties to find PHP2Shell flaws in core PHP
OpenAI says researchers at security firm Calif found the PHP2Shell vulnerabilities in core PHP using GPT-5.6 Sol as part of its Patch the Planet project.
The Decoder10.22OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon found a critical PHP flaw affecting millions of websites
OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon helped find a critical PHP vulnerability that could affect millions of websites.
Google Cloud and NVIDIA announce general availability of NVIDIA Vera Rubin infrastructure
Google Cloud and NVIDIA announced general availability of A5X instances built on NVIDIA Vera Rubin NVL72.
OpenAI models are coming to Google Cloud
OpenAI and Google Cloud are bringing OpenAI frontier models and Codex workflows to Gemini's Model Garden.
Research Papers
The Decoder10.23Anthropic's Claude now writes code and proves it safe automatically
Anthropic's Frontier Red Team, working with partner Theorem, used Claude to generate roughly 180 million lines of mathematical proof code in a few weeks.
Microsoft researchers prove the 31-year-old KLS conjecture with human-directed AI
Microsoft researcher Xinzhi Zhang and a collaborator proved the Kannan-Lovász-Simonovits (KLS) conjecture.
Microsoft and Hugging Face turn database grading into RL rewards with ThinkingBox-Training
Microsoft and Hugging Face released ThinkingBox-Training.
Tips & Opinion
Don't Worry About the Vase10.24Zvi Mowshowitz on The Hackening: AI-driven cyberattacks and the PHP crisis
Zvi Mowshowitz argues that a wave of AI-assisted cyberattacks, capped by two critical PHP vulnerabilities rated CVSSv4 9.2.