TLDR AI 2025-10-17

AI adoption is accelerating, but most teams still lack the guidance to secure, govern, and monitor it at scale.

Rob T. Lee, Chief of Research and AI Officer at SANS, developed the Secure AI Blueprint — a 3-track model to help organizations protect, utilize, and govern AI responsibly. It's a practical framework built to close the AI security gap.

⬇️ Download the blueprint for steps to defend against model poisoning, prompt injection, and adversarial attacks.

👀 Align with emerging frameworks like the Critical AI Security Guidelines

💪 Build your team's hands-on AI security skills with expert-led training

Anthropic has unveiled Claude Skills, modular task-specific packages containing instructions, scripts, and resources. These make Claude more effective at specialized tasks by loading relevant context only when needed.

OpenAI's Prompts feature within team workspaces for ChatGPT appears to be aimed at organizations. It creates a library of shared prompts made by team members accessible across the team. Team members are able to create prompt templates, set custom instructions, and integrate variables. Teams will be able to package up complex, multistep tasks into single prompts using built-in tools.

Anthropic's new Skills feature consumes only dozens of tokens per skill until needed compared to MCPs that can burn tens of thousands upfront. The design is incredibly simple: drop instructions in a folder and Claude figures out when to use them. This token efficiency and model-agnostic approach (any coding agent can read the files) will trigger a "Cambrian explosion" that makes the MCP rush look pedestrian, especially since skills transform Claude Code from a coding tool into a general automation agent capable of handling anything you could accomplish by typing commands.

GPT-5-level models now match human professionals on roughly half of economically valuable tasks, completing work in minutes that would take a human hours. OpenAI has proposed a retraining program targeting 10 million Americans by 2030, acknowledging evidence that AI is already displacing entry-level workers. The company frames AI as "more enabler than replacer" despite proposing a massive retraining initiative, which has historically failed to prevent job displacement from technological change.

If slow QA processes bottleneck you or your software engineering team and you're releasing slower because of it — you need to check out QA Wolf.

QA Wolf's fully-managed, AI-native service supports web and mobile apps, delivering 80% automated test coverage in weeks and helping teams ship 5x faster by reducing QA cycles to minutes.

With QA Wolf, Drata's team of 80+ engineers achieved 4x more test cases and 86% faster QA cycles.

⭐ Rated 4.8/5 on G2. Trusted by Cohere, AutoTrader, Salesloft, and many others.

Schedule a demo to learn more

Cognition trained specialized models that retrieve code context 10x faster than frontier models by using RL to issue up to 8 parallel tool calls per turn instead of the sequential searches typical agents use. Speed matters more than the industry realizes because it allows developers to stay in a flow state.

ScaleRL is a framework for analyzing and predicting compute scaling in LLM reinforcement learning. Their large-scale study shows that stable training recipes yield predictable performance curves, aiding efficient resource allocation.

Recursive Language Models (RLMs) can decompose and recursively interact with their input context as a variable. An RLM using GPT-5-mini outperforms GPT-5 with more than double the number of correct answers on OOLONG while being cheaper per query on average. RLMs outperform other methods ReAct + test-time indexing and retrieval over the prompt without any degradation in performance when given over 10 million tokens at inference time. RLMs trained explicitly to recursively reason are likely to be the next milestone in general-purpose inference time scaling.

Alex Lupsasca has joined OpenAI for Science, a team aiming to build AI systems to advance scientific reasoning and accelerate discovery in math and physics. GPT-5 Pro has proven its ability to tackle hard questions in science and produce answers that would take graduate students several days of research time. AI can now touch the research frontier. This is proof that AI will transform scientific research.

Manus 1.5 adds full-stack web development that handles backend infrastructure, databases, QA, and user authentication entirely, achieved entirely through natural language. The agent can test apps by itself and fix issues autonomously.

100+ million groups of images, videos, text, audio, and 3D point clouds. Sign up to the live walkthrough.

Meta's Backbone network provides wide-area network connectivity to networked locations.

Claude now connects with Microsoft 365, enabling search across SharePoint, Outlook, and Teams to surface insights and streamline decision-making within enterprise workflows.

US Uber drivers and couriers can now earn extra money by performing 'microtasks' to train AI models.

The US is estimated to deliver almost 80 gigawatts of new power generation starting in 2027.

Character.AI has released Scenes, a structured feature that lets users create interactive, narrative-driven role-play settings without needing to build a new character each time.

DeepSomatic is a convolutional neural network tool that identifies somatic variants in tumor cells more accurately across multiple sequencing platforms.