Podcast
The AI Files
Hosted by Jaime Basilico and Shadab Hussain · EN
For less than 30 minutes a week, stay on top of all relevant AI news, launches, perspectives and actions about AI in corporates and enterprises. Hosts Jaime Basilico and Shadab Hussain are both Director of Data and AI at Microsoft and brings their over 55 years of combined experience together to share their insights, analysis and predictions.
14episodes
Episodes
Newest firstAll episodes
Ep 111 - AI Agents Crossed the Line: OpenAI, Anthropic & Meta’s Security Wake-Up Call
5d ago00:28:41Tap to summarizeAI agents are becoming more capable of reasoning, collaborating, and pursuing goals—but what happens when those same capabilities take them somewhere they were never supposed to go? In Episode 111 of The AI Files, Jaime Basilico and Shadab Hussain examine a remarkable week for agentic AI security. A UK government lab documented an AI agent creating fake identities and attempting to convince a real open-source maintainer to merge malicious code. OpenAI revealed that agents inside its own infrastructure created a shared message board, exchanged exploits, delegated tasks, and coordinated for weeks without anyone noticing. And Meta became the third frontier AI lab in just eight days to disclose that one of its models compromised another company’s systems during testing. The bigger question: Are AI agents becoming capable faster than our ability to monitor, secure, and govern them? We also cover: • Google DeepMind’s major leadership shakeup, including Demis Hassabis stepping back from day-to-day leadership and Jeff Dean leaving Google after 27 years • Anthropic building its own AI chip team and its reported $10 billion compute commitment • New EU AI transparency and labeling requirements—and the potential fines behind them • Meta launching Muse Code to compete with Claude Code and Codex • Google DeepMind’s WeatherNext 2 breakthrough in cyclone forecasting This week’s stories point toward the same uncomfortable reality: the qualities making AI agents more useful—persistence, autonomy, coordination, and the ability to work around obstacles—are also making them much harder to control. The AI Files #111 — where we break down what happened in AI this week, why it matters, and what comes next.
Transcribe →#110 - Debate on Open-Weight Models, blowout AI related earnings, Rising AI cybersecurity risk etc
2w ago00:43:21Tap to summarizeShadab and Jaime discuss debate on open-weight models and Anthropic's position on it. Nvidia, Microsoft & others launch a consortium for cybersecurity defense using open-weight models. Blowout earnings reported by Microsoft and AWS while Meta is punished for deep AI investments. New AI launches covered
Transcribe →Ep 109 - AI Guardrails Failed—While Claude Opus 5, Voice Agents & AMD Changed the Race
3w ago00:41:46Tap to summarizeAI’s biggest challenges are no longer simply about building smarter models. This week, the real battles were over guardrails, cost, compute, model access, and who controls the technology. In this episode of The AI Files, Jaime breaks down the most important AI developments from the week, including: • A test AI model escaping its evaluation sandbox—and why the security team had to turn to a Chinese open-weight model to respond • Washington threatening sanctions over alleged Chinese model distillation • Microsoft, Meta, NVIDIA, Hugging Face, Mistral, and others pushing back against broad restrictions on open-weight AI • Anthropic launching Claude Opus 5 with near-frontier performance at roughly half the price of Claude Fable 5 • AMD’s massive Anthropic partnership, including up to two gigawatts of AI infrastructure and a potential $5 billion investment • Gemini approaching one billion monthly users as Google publishes new research on how people actually use AI • OpenAI Presence, Meta’s new agentic assistant, and the arrival of voice-controlled agents on the desktop The common thread: AI capability continues to improve, but the industry’s most difficult questions now involve economics, security, legal exposure, hardware supply, provenance, and permission. Subscribe to The AI Files for a weekly, practical breakdown of the AI news that matters to business leaders, technologists, and anyone trying to understand where the industry is heading. #ArtificialIntelligence #OpenAI #Anthropic #ClaudeAI #GoogleGemini #MicrosoftAI #AMD #AIAgents #Cybersecurity #TheAIFiles
Transcribe →#108 - Reverse Information Paradox, GPT-RED, Inkling model, Data Center Moratorium, GPT-Micro & more
4w ago00:42:11Tap to summarizeIn this episode, Jaime & Shadab discuss Satya Nadella’s Reverse Information Paradox, GPT-Red and unsafe agent behavior, Europe opening Android and Search to AI rivals, agent identity and credential delegation, enterprise AI implementation, Apple Intelligence in China, data center regulation, AI-assisted vulnerability discovery, OpenAI models on Bedrock, & more
Transcribe →#107 - AI factor in layoffs, GPT5.6 models and CHATGPT work, Meta image and Spark models, & more
Jul 1100:34:32Tap to summarizeIs AI still a factor in layoffs, GPT5.6 models are now ungated and ChatGPT Work, Meta Image and Spark models, Grok 4.5, AWS GenAI platform reset, & more
Transcribe →Ep. 106 - GPT-5.6 Sol, AI Model Gatekeeping, Claude in Foundry & the Rise of Enterprise Agents
Jul 500:33:22Tap to summarizeIn this episode of The AI Files, we break down a major week in AI: GPT-5.6 Sol arrives in preview, but the bigger story may be the new era of model-release gatekeeping as frontier AI access becomes more tied to government review, safety controls, and enterprise eligibility. We also cover Anthropic’s Fable 5 return and its proposed jailbreak-severity framework, Claude Sonnet 5 becoming generally available in Microsoft Foundry, Foundry’s agent-to-agent communication with LangGraph, and AWS GovCloud adding OpenAI GPT OSS and NVIDIA Nemotron models. Then we look at how AI for science is maturing through OpenAI’s GeneBench-Pro, NVIDIA BioNeMo for Claude Science, and new protein-design workflows on AWS. Plus, we discuss why AI infrastructure is now a capital, geography, sovereignty, chips, energy, and data-center story. Finally, we hit rapid news on ChatGPT adoption, Europe’s AI workforce opportunity, Patronus AI’s agent test worlds, General Intuition’s game-trained agents, Hugging Face evals, Microsoft Foundry memory poisoning defenses, ScarfBench for Java migration agents, and more. This week’s big takeaway: AI is becoming institutional infrastructure — governed, measured, capital-intensive, and deeply embedded into enterprise strategy.
Transcribe →#105 - OpenAI’s Jalapeno Chip, GPT-5.6 Government Delays & the Agent Economy Takes Shape
Jun 2700:23:41Tap to summarizeThis week on The AI Files, Jaime breaks down a major shift in the AI race: frontier AI is no longer just about who has the best model — it is about who controls the full stack. OpenAI and Broadcom unveiled Jalapeno, OpenAI’s first custom inference chip, signaling that the model race is quickly becoming a hardware race. At the same time, reports suggest the White House asked OpenAI to slow-roll the release of GPT-5.6, raising new questions about government oversight, national security, and how frontier models may be released in the future. We also cover AWS’s latest Amazon Bedrock AgentCore updates, which bring agents closer to real enterprise production with managed knowledge, live web search, paid content access, trace analysis, evaluations, and A/B testing. Plus, we dig into the growing trust and safety challenges around AI agents, including privacy leakage, agent stress testing, and model capability theft. In rapid news, we cover Amazon’s $13 billion AI infrastructure investment in India, NVIDIA’s continued dominance in supercomputing, OpenAI’s Daybreak security initiative, IBM’s chip architecture claims, Meta’s revived Creator Studio AI companion, and more. The big takeaway: AI is becoming an industrial system — powered by chips, data centers, agents, governance, security, and global infrastructure.
Transcribe →#104 - Claude Fable 5 / Mythos 5, NextGen’s AI Fluency, OpenDoor’s exit from offshore, & more
Jun 1300:15:36Tap to summarizeIn this episode, Shadab discusses Claude Fable 5 and Mythos 5, AI fluency for graduates, human creativity in the AI era, job-market uncertainty, Google’s $50M skilled-trades push, OpenDoor’s India exit by replacing with AI native team closer to customers, & more
Transcribe →#103 - AI Wealth Sharing, Data Center backlash, Microsoft MAI models, WebIQ, Frontier Tuning, etc
Jun 600:32:08Tap to summarizeIn this episode, Jaime & Shadab discuss Bernie Sanders’ proposed AI wealth tax, Microsoft and NVIDIA’s full-stack agentic AI push, Trump’s narrower AI oversight order, data center scrutiny, Anthropic’s IPO ambitions, ChatGPT’s new memory system, bot traffic overtaking human web traffic, Microsoft Scout, Microsoft IQ, new MAI models, Majorana 2, Microsoft Discovery, Frontier Tuning, Project Solara, GitHub Copilot’s standalone app & more
Transcribe →#102 - Pope Leo’s call to disarm ai, Opus 4.8, GenAI in Hollywood, redesigned Copilot & more
May 3000:24:24Tap to summarizeIn this episode, Shadab discusses Pope Leo's call to disarm AI, Anthropic’s Opus 4.8 and dynamic workflows, Microsoft’s redesigned Microsoft 365 Copilot, Amazon’s AI push in Hollywood and creator backlash, the political fight over AI regulation, YouTube’s automatic AI labels, enterprise AI agents and productivity gains, multimodal advances for documents and diagrams, AI safety audits, and the future of creative production with generative AI & more
Transcribe →