Video by Open Data Science and AI Conference via YouTube
This week, OpenAI disclosed two cases of agents weakening their own safety monitoring, an outside audit found gaps in an earlier version of its monitoring system, and the company moved frontier model training offline. Two urgent security stories followed: fixes for critical PHP flaws and a compromised LiteLLM release that stole credentials. In tools and research, Anthropic previewed Claude Duet, Google Research released Harness2, and Cloudflare cut the cost of its AI code reviewer by 68%. Anthropic changed its API, DeepSeek released the DeepInfer inference engine, and Google Cloud now carries both Claude and OpenAI models. In jobs, Indeed found AI keywords in 7.3% of US postings. The episode also covers the DHS order on DeepSeek, Anthropic’s reported S-1, and a round of quick hits.
In this episode
OpenAI’s agent monitoring incidents, external audit, and offline frontier training.
Critical PHP2Shell vulnerabilities and the LiteLLM supply chain attack.
Anthropic’s Claude Duet and Google’s Harness2 framework.
Cloudflare’s 68% AI code reviewer cost reduction.
Anthropic’s API changes and DeepSeek’s DeepInfer engine.
Google Cloud expands access to Claude and OpenAI models.
AI keywords appear in 7.3% of US job postings.
DHS orders agencies to remove DeepSeek; Anthropic’s reported S-1 reveals financial details.
Quick hits: NVIDIA OpenShell, Kimi K3.1 Pro, Recursive Flint, biology agents, AI salary premiums, and NYC’s AI Jobs Institute.
References & Resources
OpenAI Agent Monitoring — ABC News
Agent Monitoring Audit — METR
Frontier Training Security — OpenAI
LiteLLM Security Incident — BleepingComputer
Claude Duet — Anthropic
Harness2 — GitHub
Inference Optimization — Cloudflare
DeepInfer — GitHub
AI Job Market — Indeed
DeepSeek Federal Order — TechCrunch
This episode was sponsored by:
ODSC AI West 2026 returns October 27 to 29 at the Hyatt Regency San Francisco Airport in Burlingame, with more than 250 speakers across hands-on training in machine learning, LLMs, agentic AI, and AI engineering. Learn more: odsc.ai/west
ODSC AI NYC – AI for Work is a hands-on event taking place December 2–3 in Manhattan. Through practical sessions and workshops, attendees will learn how to build and apply AI. Learn more: https://summit.ai
ODSC AI East 2027 – The Leading AI Builders Conference. Join us at the Boston Convention Center from May 10–12, 2027 for expert-led sessions on AI Engineering, Data Science, Machine Learning, Agentic AI, and AI-driven automation. Learn more: https://odsc.ai/east
ODSC AI Engineering Accelerator – From Using AI to Building It. Seven weeks of live, cohort-based training in LLMs, AI assistants, agents, and AI engineering. Ship a working AI agent or assistant by the end. Winter and Spring cohorts available. Learn more: https://odsc.ai/accelerator
To watch more videos like this, visit https://aiplus.training
Sign up for the newsletter to stay up to date with the latest trends in data science: https://opendatascience.com/newsletter/
Follow us online!
• Facebook: https://www.facebook.com/OPENDATASCI
• Instagram: https://www.instagram.com/odsc/
• Blog: https://opendatascience.com/
• LinkedIn: https://www.linkedin.com/company/open-data-science/
• X (twitter): https://x.com/_odsc