Anthropic says Project Glasswing partners have already used Claude Mythos to surface more than 10,000 high- or critical-severity vulnerabilities in essential software. That number makes the AI story look less like benchmark theater and more like an output layer you can measure in tickets, triage queues, and remediation backlog.

What matters is not just that a model found more bugs. It is that capability is starting to be read through security throughput: can a model expose dangerous work fast enough to change where the bottleneck sits inside security operations? If the answer is yes, AI advantage will increasingly be measured by how much real output a system can push into workflow, not just how impressive it looks in a sandbox.

Tibo says about 5% of OpenAI production traffic now sits on the Pi harness, with another 5% on OpenCode, while also reminding users that they can bring their ChatGPT account into other tools. Because this is a product-leader disclosure on X, it needs to stay framed as a claim-by-source, not as a complete market census.

Frontier demand is no longer locked inside the first-party shell. Once real traffic starts flowing through outside harnesses, the race is no longer just about who has the better model. It becomes about who can place that model into more execution surfaces while still keeping usage.

The Information says Google is pitching Gemini 3.5 Flash as a lever that could save very large cloud customers more than $1B a year versus rival frontier models. Because the underlying article remained paywalled in this run, this also needs to stay a claim-by-source, even with VentureBeat supplying open context on how Google is packaging the cost story around 3.5 Flash.

Google is trying to drag the model race out of the “who is smarter” frame and into procurement and margin. Once buyers start reading frontier AI through the invoice instead of the demo, competitive advantage shifts toward whoever can turn a model into the clearest economic weapon for the enterprise.

OpenAI says Gartner now places it in the Leader quadrant for enterprise coding agents: OpenAI says Gartner placed it in the Leader group as the market for enterprise coding agents enters a new phase of expansion and competitive realignment. The signal is that coding agents are now being formalized as a software category with actual vendor sorting, not just as an extension of chat.

Google says Managed Agents and Antigravity turn agent workflows into a managed runtime with remote Linux sandboxes: Google says Managed Agents can spin up remote Linux environments, browser access, tools, and orchestration through Antigravity. The moat in agents is moving quickly toward runtime and execution control, not just the model endpoint.

Anthropic says KPMG is embedding Claude into Digital Gateway for 276,000 employees: Anthropic says KPMG is bringing Claude into Digital Gateway and expanding access across 276,000 employees, starting with tax and legal workflows. That matters because it shows AI getting pulled directly into governed workflow surfaces, not just added as another seat count.

Warp says teams can now bring their own inference endpoints into its agent stack: Warp says teams can route its agent stack through custom or self-hosted inference endpoints instead of sticking to model defaults. If that trend holds, the most valuable control surface may sit in routing and portability more than in the model UI itself.

The Information says Anthropic and OpenAI spend is helping buyers force tougher contract terms on legacy software vendors: The Information says rising Anthropic and OpenAI spend is giving enterprise buyers more leverage to pull budget away from traditional vendors through shorter and tougher contracts. AI spend is no longer just a new cost line. It is starting to rewrite pricing power in the old software stack.

Bloomberg says Anthropic could close a new $30 billion-plus round at a valuation above $900 billion as soon as next week: Anthropic is nearing a new round that could top $30B at a valuation above $900B, according to people familiar with the matter. If the round closes as currently reported, it would show how aggressively capital markets are still rewarding frontier AI scale and momentum.

The practical lesson today is not “how strong Codex is.” It is how Codex can be used as an embeddable component. Dominik Kundel says developers can point their app at Codex app servers, keep the stack open source, and support sign in with ChatGPT.

For teams thinking about AI product surfaces, that is an important shift: instead of building everything from scratch or sending users to a separate tool, you can test a middle lane where Codex becomes the capability layer inside your own app.

If you spend 30 minutes on this today, the goal is not to build a whole new product. The goal is to identify which user job inside your app is actually worth turning into a first Codex-backed surface.

Do this now:

Tiny Scorecard:

How to read it:

That’s 6 sales in 7 months. 29 all time. And the performance?

16.5%, 17.6%, and 17.8%, net annualized returns on sold works held longer than one year (See all 29 at Masterworks.com)

It’s not from stocks, private equity, or real estate… it’s from contemporary and post war art. Crazy, right?

With Masterworks, you don’t need to be a BILLIONAIRE to invest in multi-million dollar art anymore.

Historically, the segment overall has had attractive appreciation and low correlation to stocks.*

Masterworks targets works featuring legends like Banksy, Basquiat, and Picasso, identifying what they believe to have significant long-term appreciation potential, not just at the artist level but at the level of individual artworks.

As one of the largest players in the art market, with $1.3 billion invested over 500 artworks, they pass critical advantages through to their 70,000+ members to add art to their portfolios strategically.

Looking to diversify your investments in 2026?

Click here to skip the waitlist

^{*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.}

TechCrunch turns Google’s latest launch cluster into a clear thesis: the company is trying to move gravity away from chatbot framing and toward agentic product surfaces and runtime infrastructure. Read it if you want to see the Google cluster today as a platform move, not an event-day recap.

Gartner reads coding agents as a category that is being formalized quickly, with clearer vendor sorting and buyer expectations. This is the reading to keep if you want to understand why coding agents now deserve to be read as a real software market, not just as another model UX trend.

The WSJ pushes AI coding out of the “ship faster” frame and into code quality, maintainability, and trust. It adds the necessary counterforce to today’s issue: if AI speeds up shipping but leaves slop behind in production, the bottleneck eventually snaps back to review, debugging, and trust between teams.

ITPro reads the current AI spend cycle as a force that is pushing incumbents to rethink pricing and contract logic just to stay relevant. It ties Section 4 back to Section 2 by showing that AI is not only creating new winners at the model layer, but also grinding away at the economics of old software.