Mike Vizard, Jon Swartz, Fred Wilmot, and Jack Poller break down how Anthropic's own safety testing let Claude models breach three real organizations, a new survey on wasted enterprise AI spending, and IBM's report on the 56% surge in AI-enabled cyberattacks.

Mark Zuckerberg gave four rare interviews this week arguing against concentrating AI power — even as Techstrong.ai's Alan Shimel notes Meta's own new Muse Spark model is closed and now sold commercially. A new Futurum Group survey of 839 IT decision-makers finds 75% of organizations have hit a production issue caused by AI coding tools, with Futurum's Mitch Ashley warning vendors "will lose the renewal conversation" if they can't prove value. And Alan Shimel makes the case that cloud native didn't get replaced by AI — it got inherited by it. Full breakdown on today's Techstrong Gang with hosts Alan Shimel and Mike Vizard, plus Jon Swartz, Mitch Ashley, and Robert Reeves.

NVIDIA is in talks to guarantee up to $250 billion in financing for OpenAI's Ohio data center campus, deepening a financial web that Techstrong.ai's Alan Shimel says makes OpenAI "too interconnected to fail." Samsung landed a $200 billion-plus Broadcom partnership through 2030 spanning memory and foundry manufacturing. And a humanoid AI teacher's assistant was suspended from a New York school district after locals learned about its maker's past. Full breakdown on today's Techstrong Gang with host Alan Shimel, Mike Vizard, Jon Swartz, Elizabeth Safran, Teri Robinson and Jack Gold.

Model Context Protocol shipped its biggest revision since launch, going fully stateless so MCP servers can run behind standard load balancers like ordinary microservices. A rogue OpenAI model hacked into Hugging Face, prompting Nvidia to rally 30+ companies into a new Open Secure AI Alliance. And panelist Chris Blask argues cybersecurity should shift from protecting components to stewarding relationships between them. Full breakdown on today's Techstrong Gang with host Alan Shimel, Mike Vizard, Chris Blask, and Kate Scarcella.

Anthropic cut 80% of Claude Code's system prompt with no drop in coding performance, moving toward leaner, on-demand context for AI agents. US software developer employment for ages 22-25 has fallen more than 20% since ChatGPT launched, as AI coding agents take over the routine work junior developers used to learn on. And Microsoft launched two in-house AI models, cutting its reliance on OpenAI. Full breakdown on today's Techstrong Gang with host Alan Shimel, Mike Vizard, Jon Swartz, Stephen Foskett, and Garima Bajpai.

Linux kernel maintainers published 432 CVEs in two days, and security teams are calling the volume unmanageable. The European Union fined Google $1 billion for violating antitrust rules tied to search and the Google Play Store. And Samsung is negotiating a stake in AI startup Mistral at a $22.8 billion valuation. Full breakdown on today's Techstrong Gang with host Jon Swartz, Tracy Ragan, Jeff Reich, and Jack Poller.

OpenAI's own models broke out of a security-testing sandbox and hacked into rival Hugging Face's servers using stolen credentials. Alphabet answered the AI security conversation with three new Gemini Flash models, including one built to hunt and patch vulnerabilities. And AMD signed a multibillion-dollar chip and investment deal with Anthropic while Microsoft adopted AMD's Helios system for Azure. Full breakdown on today's Techstrong Gang with Alan Shimel, Jon Swartz, Jack Poller, and Tom Hollingsworth.

Alan Shimel makes the case that Google has become a "parasite" on the open web, consuming publisher content through AI Overviews while eliminating the reason to click through. A federal judge grants final approval to a $1.5 billion Anthropic copyright settlement — the largest in U.S. history. And Forrester's 2027 forecast finds tech and marketing budgets surging as AI pricing shifts from per-seat to usage-based models. Full breakdown on today's Techstrong Gang with Alan Shimel, Jon Swartz, Teri Robinson, and JP Morgenthal.

Alan Shimel makes the case that the AI model war ends in abundance, not victory, as open and controllable models undercut proprietary AI's pricing power. Cisco open-sources Antares, ultra-cheap AI models purpose-built to hunt software vulnerabilities. And Anaconda acquires Kilo Code, pushing the Python company deep into agentic AI development.

Full breakdown on today's Techstrong Gang with Alan Shimel, Kate Scarcella, Sid Nag, and Chris Blask.

Today's Techstrong Gang panel — Alan Shimel, Jon Swartz, Mitch Ashley, Yvette Schmitter, and Liz Safran — breaks down why being "indispensable" doesn't protect any layer of the AI stack from commoditization, Mira Murati's Thinking Machines taking on Chinese open-weight dominance with Inkling, China's looming Kimi K3 release, and OpenAI's GPT-Red model beating human red-teamers 84% to 13% while making GPT-5.6 six times more resistant to prompt injection attacks.