BIG TECH

OpenAI and Anthropic near a scary leap forward

OpenAI and Anthropic are about to launch new AI models for a world changed by OpenClaw. 

Anthropic has its "Claude Mythos" model, and OpenAI is preparing to launch its "Spud" model. Both companies are touting these as extremely powerful models that represent the next big leap for LLMs. And if that's the case, despite the codenames, these models could roll out as Claude 5.0 and ChatGPT 6.0. 

There are several factors to watch:

• OpenAI CEO Sam Altman said the new Spud model is so powerful that it will "really accelerate the economy"

• Anthropic CEO Dario Modei said that the company is already testing Claude Mythos with early access customers

• A leaked Anthropic blog post obtained by Fortune stated that the Anthropic team believes its model is so powerful that it could pose unprecedented cybersecurity risks, and that's one of the reasons why Anthropic has been warning government officials and giving organizations early access to the model in order to help them prepare

• OpenAI's Spud could also be the foundation of the company's new "superapp" that will reportedly combine the desktop apps for ChatGPT, Codex (its coding tool), and Atlas (its web browser) into one streamlined experience with advanced AI agent capabilities 

• As another nod to how powerful OpenAI believes its next model will be, the company reportedly renamed the product organization led CEO of Apps Fidji Simo to the "AGI Deployment" team

One reason these models are advancing so rapidly and reaching a higher stage of development is that the labs are now using the models themselves to help build and improve the models. Sam Altman mentioned this when the company released GPT-5.3-Codex. This is a process called recursive self‑improvement (RSI), and it has long been anticipated as the point where AI systems will make a dramatic leap forward.

We've already seen AI hit a new gear in 2026 with personal AI agents unleashing a wave of automation, productivity gains, and software improvements. That was led, of course, by OpenClaw. And now we're seeing the benefits OpenClaw has unlocked being adopted by AI and software companies across the tech industry. That agentic movement is about to get supercharged by these new models. But as we give the agents more rocket fuel, it's important to remember that they're non-deterministic systems that can go rogue and do things we didn't expect. That's going to make it more important than ever that we build good harnesses and systems to contain these agents to protect against the worst possible outcomes.

TOGETHER WITH SNYK

Building an AI Security Champions Program

Join Snyk Staff Developer Associate, Sonya Moisset, on April 9 at 11 AM ET for a live workshop: Building an AI Security Champions Program.

Learn how to:

• Adapt the champion role for AI-driven workflows

• Enable developers to secure AI-generated code

• Define success with real KPIs

Plus, earn 1 ISC2 CPE credit. Save your spot.

BIG TECH

Google Veo pushes video AI forward, cuts prices

OpenAI may have shed Sora, but Google’s AI video ambitions are far from over. 

On Tuesday, the company announced Veo 3.1 Lite, the latest edition to its video generation family of models and its “most cost-effective video model,” the company said in its announcement. Google said that Veo can build high-volume video applications at half the cost and roughly the same speed as its previous Veo model. 

Veo 3.1 Lite supports both text and image inputs, and can generate both landscape and portrait ratios at resolutions up to 1080p. Veo 3.1 Lite can cost as little as 5 cents per second, compared to 40 cents per second for Veo 3.1 standard. The model is available to developers in both the Gemini API and Google AI Studio. 

Additionally, Google is giving users a discount on Veo 3.1 Fast, its mid-range video generation model, starting April 7, cutting generation costs to 10 cents per second for 720p and 12 cents per second for 1080p. 

In a post on X, Logan Kilpatrick, a member of technical staff for Google DeepMind, said that video is “here to stay.” 

Google’s courting of AI video customers comes as OpenAI casts off its own video-generation efforts by ditching Sora and ending its $1 billion, three-year licensing partnership with Disney. The company’s Sora switchup signals a broader refocusing of its compute towards more revenue-generating products and redirecting its video model staff to world models. Since video models are incredibly compute-intensive, it makes sense that the video model was on the chopping block. 

“We cannot miss this moment because we are distracted by side quests,” Fidji Simo, CEO of applications at OpenAI, told staff in an all-hands meeting, according to Business Insider.

But Google isn’t the only one trying to fill the vacuum left in Sora’s wake. Elon Musk-owned xAI is “doubling down” on AI video with the next release of Grok Imagine, the lab’s own video generation model. But as xAI faces an ever-growing pile of lawsuits over its image generation capabilities, Google may have a better shot at appealing to a wider audience.

Google’s ability to press further into video generation while OpenAI backs off is a prime example of the company’s edge in the AI market. While AI companies scrap for every morsel of compute they can get their hands on, Google sits atop a legacy. The company spent years building Google Cloud into one of the big three cloud hyperscalers, giving it a much-needed wealth of compute resources. Additionally, having cemented its billions in the digital advertising and search markets, Google doesn’t have to worry about the revenue picture as much as rivals like OpenAI and Anthropic. Because of this, Google can take on the “side quests” as Simo called them, cornering a market while it’s still nascent. 

TOGETHER WITH VIKTOR

Forget single-purpose AI agents, Viktor gives you a full-blown AI peer.

Viktor is an AI coworker that lives in your Slack or Microsoft Teams. No hardware, no setup, enterprise security from day one. It connects to 3,000+ tools (Stripe, HubSpot, Meta Ads, GitHub, Notion), delivers board-ready PDFs, builds web apps, and runs scheduled tasks 24/7. SOC 2 certified. Your data never trains models. Watch Viktor work ->

Backed by Daniel Gross (Y Combinator), Nat Friedman (ex-GitHub CEO), and Mati Staniszewski (ElevenLabs). See why ->

Give Viktor your worst task (for free) ->

PRODUCT

Slack adds AI meeting notes and a desktop AI agent

For years, Slack has been where teams go to communicate. Now, it's setting its sights on something bigger: replacing your favorite AI apps without you ever leaving the platform.

On Tuesday, Slack announced more than 30 new capabilities that not only expand what its Slackbot AI agent can do but also introduce new features previously unavailable in Slack. For instance, one of the biggest highlights is a new meeting transcription and note-taking feature that acts much like Otter.ai or Fireflies.ai. 

“For sales teams, the personal note taker is huge, because they often have their email open, the Zoom meeting open, and half paying attention because they're working on something for the meeting that they're in, so I think that's immensely powerful,” Rob Seaman, EVP & GM, Slack, told The Deep View. 

To use the feature, users just have to have the Slack desktop app installed. Then, when activated, Slackbot can listen in the background of your meetings when activated to capture the content, take notes, produce summaries, identify action items and more. The best part is that the meeting can take place in other apps, such as Zoom or Google Meet. Some other new features include: 

• Operator mode: Slackbot can complete multi-step tasks across your desktop app on your behalf. All users have to do is select anything on their screen and ask Slackbot to act on it, such as "Summarize this” or “Draft a follow-up.” 

• AI skills: Slackbot can access a built-in library of AI skills or instructions on how to do something, as well as user-built ones, for repeatable, shareable workflows. 

• MCP: The new MCP client allows Slackbot to access any tool, including enterprise apps such as Google Workspace, Microsoft 365, Notion, and more. 

• CRM: Slack now has a native CRM, in which Slackbot can read channels to log deals, contacts, and call notes. This is meant to help small businesses that lack a CRM or a sales department.  

All of the new features are meant to build on the app’s existing AI agents, including Slackbot, which, the company says, is on track to become the fastest-adopted feature in Salesforce history, less than two months after its January launch, when it was redesigned as a deeply personal AI agent for work. Slackbot will be available to all Free and Pro Slack users starting in April. 

"We're building a personal agent [Slackbot] that's got all of your context in Slack, but is immensely capable, and this is like a highly intelligent teammate that can help you get your work done throughout the day, but also interface with those other agents,” added Seaman.

One of the biggest bottlenecks in workflow optimization is the need to constantly context-switch between different apps to get things done. However, most companies already rely on a single communications app, constantly checked for collaboration, as their central hub. If that hub could also provide access to other tools, such as a transcription app, it would help solve the context-switching problem. That's exactly what Slack has done. This is an example of enterprise software that delivers genuine value to users rather than just AI-washing about new add-on features. It's a trend I hope is here to stay.

LINKS

• Iran to target major US tech firms in middle east, including Apple, Google

• OpenAI boosts historic funding round to $122 billion at $852 billion valuation

• Alibaba’s new model is proprietary, pivoting away from open-source

• Runway launches $10 million fund to support AI startups

• California’s Gov. Gavin Newsom issued an executive order on AI

• Northwestern University study shows federal judges use AI in judicial work

• CoreWeave Inc. raised $8.5 billion for cloud computing capacity expansion

• Codex: OpenAI introduced a Codex plugin for inside Claude Code

• Microsoft Copilot: Council lets users run multiple models on same prompt simultaneously 

• Speechify Voice AI for Windows: Now available to Microsoft Windows users

• Perplexity API Platform: The AI search engine’s API now covers the full agent stack

• Scale AI: Research Scientist, Agent Robustness

• Affirm: Senior Staff Machine Learning Engineer, (ML Underwriting)

• Palo Alto Networks: Principal Machine Learning Engineer

• Amazon: Applied Scientist, LLM Code Agents, Kiro Science

GAMES

POLL RESULTS

Do you think the Americans public's opinion of AI will broadly be favorable by the end of the decade?

Yes (42%)
No (49%)
Other (9 %)