PRODUCTS

Anthropic's Mythos launch is a layered compromise

Anthropic has finally released the world's scariest language model, if you believe its warnings. 

On Tuesday, the AI firm launched two "Mythos-class" models, Fable 5, which it claims is safe for general use, and Mythos 5, which has safeguards lifted in some areas. The company said that both models are state-of-the-art on nearly all tested AI benchmarks and can work autonomously for longer periods, exceeding the capabilities of any model Anthropic has ever made generally available. 

Here are the details: 

• Fable 5 provides exceptional performance in domains such as software engineering, knowledge work, vision and scientific research, and is capable of handling longer and more complex tasks than previous iterations. However, given that this version of the model is available to the general public, queries on certain topics will be sent to Claude Opus 4.8, Anthropic's next most capable model, released at the end of May. 

• Mythos 5, meanwhile, doesn't have such safeguards. This model is available only to a small group of cyberdefenders and infrastructure providers and will still be deployed through Project Glasswing. Anthropic said that Mythos 5 has the strongest cybersecurity capabilities of any model in the world, and that it intends to expand access through a "broader trust program." 

• Anthropic is offering both models at $10 per million input tokens and $50 per million output tokens, half the cost of the Mythos preview. Until June 22, Fable 5 will be included on Claude Pro, Max, Team, and seat-based Enterprise plans at no extra cost. After that, using it after that will require usage credits.

Several early testers of Fable, including Cursor, GitHub, and Lovable, broadly confirm that the model is the most capable in the Anthropic lineup they've tested. Figma Director of Developer Product Matt Colyer said that the model is a "clear step forward in agentic coding and prototyping." 

Additionally, Anthropic announced changes to its data retention policy for business customers. For Fable, Mythos and all future releases with similar or higher capabilities, Anthropic will require 30-day retention for all usage in both first- and third-party services. Though Anthropic says it will not train the models on this data, using it only to defend against "complex and novel attacks," it's a stark shift from the company's previous Zero Data Retention policy. 

The release follows the company's announcement of a $65 billion funding round at a $965 billion valuation, which it said would allow it to significantly expand its compute capacity to scale Claude models.

It's an interesting time for Anthropic to release its most capable model yet, given that just last week, its researchers called for an industry-wide slowdown on model development amid concerns that recursive self-improvement could arrive sooner than expected. However, it's inevitable that the larger these companies get, the more beholden they become to shareholder interests. Given that the company is preparing for an IPO at the same time as rival OpenAI, it may have believed it had no choice but to release Mythos, even in a somewhat stunted capacity as Fable. And now, backed by a pile of cash that allows it to actually scale these compute-hungry Mythos-class models, Anthropic has the resources to release them. 

TOGETHER WITH IBM

From Java 11 to 21 in three days

Blue Pearl’s platform had grown over several years and needed an update to deal with deprecated APIs, aging dependencies and limited automated testing. The codebase included 127 deprecated API calls, more than 50 outdated libraries and known CVEs, which made a traditional upgrade a 30-plus-day effort.

Working with IBM® Bob as a programming partner, the team completed analysis, refactoring and test creation in three days. The platform moved to Java 21 LTS with 92% test coverage, zero CVE-bearing dependencies, no production incidents and estimated 15% JVM performance gain.

Read the full story

BIG TECH

Apple's killer AI feature is the one we ignore

Apple's most powerful asset in AI is also the easiest to overlook and take for granted.

When Apple rebooted Apple Intelligence and Siri AI on Monday at WWDC 2026, nearly all the features announced were playing catch-up with the AI labs and the phone makers (with the one big exception being Spatial Reframing, which is quite impressive and unique). 

But there are three ways that even Apple's me-too features can provide advantages for a lot of users and bring new people into AI:

• Usability: Apple has become Apple by making complex things simple. And AI can get incredibly complex. One of the ways Apple is removing friction in AI is by disaggregating features and integrating them into the places where we're most likely to use them, rather than forcing us to always open a chatbot window. In other words, on the home screen, in the keyboard, in the camera app, in the Shortcuts app, etc. 

• Privacy and safety: While other AI and device makers are deeply focused on running AI in the cloud, Apple prefers to run its AI either on-device or on its Private Cloud Compute platform, ensuring that even Apple itself can't view, access, or misuse your most sensitive data. 

• Personal Context: Apple can tailor AI to answer questions or perform actions based on information stored on your Apple devices, including email, text messages, notes, photos and documents. That can be enormously useful, but also be scary since it's highly sensitive information. So it requires deep trust that stems from strong policies and frameworks around security and privacy.  

All three of these factors work together, but it's Personal Context that is ultimately Apple's greatest superpower in AI. 

Consider this: when OpenClaw burst onto the scene earlier this year, it very rapidly became the most useful personal AI agent that anyone had seen. And even though it was difficult to configure and risky to run, it became a cultural phenomenon in the tech industry, with over half a million people excited to use it. Why? Because OpenClaw users gave it access to a whole computer, which typically contained a treasure trove of context, including messages and documents. 

OpenClaw wasn't the first personal AI agent, and it wasn't even necessarily the most advanced. It was just the one that operated off the most context. And as a result, it was able to do more and create more value for its users. 

Apple's Personal Context has the potential to do the same. It doesn't need to have the most bleeding-edge features or the most powerful models, as long as it has strong enough AI capabilities and enough trust that users will be confident giving it access to more of their data and context.

Apple's closest competitor when it comes to context is Google's Personal Intelligence. The difference is that Google's service is centered around cloud data and your information stored in Gmail, Google Photos, Search, Gemini memories, and other Google services. Because Google makes money from selling ads against the information it knows about you, many people are naturally less willing to trust Google to protect their privacy. That trust is something Apple still retains with most of its users, despite its slow progress in AI. And if the new Siri AI turns into a hit, it will likely be because people trust Apple enough to take advantage of everything it can unlock with AI and Personal Context.

TOGETHER WITH WALL STREET PREP

AI fluency is the new business fluency.

Columbia Business School Exec Ed + Wall Street Prep's 8-week AI for Business & Finance Certificate Program was built specifically for non-technical roles.

You’ll learn just enough Python to develop AI skills you can apply right away to enhance productivity and business impact.

• Build AI workflows you can apply to financial analysis, modeling, and reporting

• Join LIVE weekly office hours with Columbia faculty

• Earn a Columbia certificate and join a 5,000+ network of finance and business professionals

Code DEEPVIEW takes $300 off tuition + $200 more for early enrollment by June 15.

CONSUMER

Where Apple's AI strategy runs deeper than Siri

A revamped Siri stole the spotlight at Apple's Worldwide Developers Conference, but it wasn't the only headline. Deep Apple Intelligence integrations across the company's entire product ecosystem proved just as compelling, if not more so.

Like Siri, Apple Intelligence also benefits from the latest Apple Foundation Models (with help from Gemini), making it smarter and more capable and enabling it to fuel new experiences across Photos, Safari, Passwords, Shortcuts, Image Playgrounds and more. The new features all focus on enhancing everyday tasks for users. 

My favorite and the most unique feature is Spatial Reframing, which lets users recompose photos by shifting the subject's perspective with a touch and drag, generating new content to fill the gaps (it's better understood visually, so check out this clip). Other new photo-editing additions include Extend for expanding images to fix the framing, and an upgraded Clean Up tool with a new high-quality mode for difficult edits.

Safari also received what I thought were standout upgrades, highlighted by two Apple Intelligence-powered features. Tab grouping automatically organizes open tabs by topic, while Notify Me lets users set natural language alerts for website changes: from price drops to something as specific as a new ice cream flavor appearing on a menu. 

Without explicitly saying the word "agent," Apple also released two new features that accomplish some agentic things: extensions and shortcut automation. People can now create extensions in Safari by describing what they would want it to do. Similarly, in Shortcuts, users can now describe a Shortcut to automate tasks, with Shortcuts figuring out how to assemble the steps for the user. 

You can find a rapid-fire list of more features below: 

• Passwords: It can now automatically fix compromised passwords for users with just a tap, and even handle multiple at the same time. 

• Image Playground: The image generator, now powered by a new Private Cloud Compute model, produces high-quality images in a new photorealistic style, and easier editing via tap, circle, or highlight.

• Messages: Users get one-tap suggestions based on the context of conversations.

• Phone: Call Context surfaces relevant information, such as a reservation number.

• Home Kit: Users can now receive a single summary of multiple notifications and read a summary of video descriptions from cameras.

After so much anticipation, Siri unsurprisingly stole the show at WWDC, but the real standouts were some of the Apple Intelligence features that went under the radar. These aren't novelties. They are things people in the Apple ecosystem could use almost every day. Personally, I've long since stopped using Siri, finding it more frustrating than helpful. While I look forward to trying the new Siri, it will take some getting used to. The Apple Intelligence updates, on the other hand, deliver tangible value by upgrading the tools I rely on constantly. When packaged with Siri, users will be able to take full advantage of a broadly upgraded Apple AI experience.

LINKS

• Standard Bots raises $200 million to bring robotics manufacturing to US

• Perplexity is planning a 2028 IPO, CEO Aravind Srinivas said

• Microsoft cut hundreds of Chinese employees in Azure unit 

• Sam Altman says OpenAI is entering "third phase," aiming for personal AGI

• More than 34,000 Instagram accounts compromised due to Meta AI tools

• Gopuff, xAI partner on in-app, agentic personal shopping tool

• Gemini 3.5 Flash Live Translate: Google's new real-time speech-to-speech translation model, with support for more than 70 languages. 

• North Mini Code: Cohere's first open source, agentic coding model, lightweight and efficient at just 30 billion total parameters with 3 billion active. 

• Luma Ray3.2: Luma's latest video model for cinematic editing, with better control and continuity than previous models. 

• Canva in ChatGPT: Users can now turn OpenAI-generated images into editable Canva designs, without leaving chat

• AMD: Principal GenAI Inference Optimization Engineer

• Unconventional AI: AI Systems, Language & Reasoning Models

• Anthropic: Research Engineer/Research Scientist, Pre-training

• Faraday Future: Tech Lead, Robotic AI Model

GAMES

A QUICK POLL BEFORE YOU GO