In partnership with
Good morning. It’s Friday, May 1st.
The Musk–Altman trial is turning into the AI industry’s founding myth on cross-examination.
Musk says OpenAI betrayed its nonprofit mission and became the kind of profit-driven company it was created to counter. OpenAI says Musk knew the structure.
X user NIK is following the trial closely and doing some excellent reporting if you want to follow along!
-Jeff
AI Breakfast
You read. We listen. Let us know what you think by replying to this email.
Why The 'AI Reset' of 2026 Could Blindside Most Investors
Most investors think the AI boom is already underway. But one 47-year Wall Street veteran says everything we've seen so far is just phase one. A new wave of AI infrastructure is coming online in 2026. When it does, it could completely reshape which companies lead the next cycle — and which ones get left behind. He's now recommending specific stocks to buy ahead of what he calls "the AI Reset," and the names to exit before it hits. The investors who understand what's coming could be positioned for the biggest gains of the decade. The ones who don't may not get a second chance. Get his full list of buys and sells here — no cost, no obligation.
Claude’s “Jupiter” model appears ahead of Anthropic’s developer event
A new Claude model appears to be moving through Anthropic’s release pipeline.
The model, reportedly labeled claude-jupiter-v1-p, has surfaced in red-team testing ahead of Anthropic’s Code with Claude developer conference on May 6. TestingCatalog says Anthropic has begun internal red teaming for the model, which usually means a release candidate is being tested for safety, reliability, and misuse before a broader launch.
That timing is hard to ignore. Anthropic’s developer conference is built around hands-on workshops, live demos, and new Claude capabilities, which makes Jupiter look less like a random internal test and more like a possible launch candidate.
Anthropic has already been pushing Claude deeper into coding and agentic workflows. Its latest release, Claude Opus 4.7, is described by the company as a smarter model for coding, agents, vision, and complex professional work.
The obvious read: Jupiter may be another coding-focused Claude release, possibly a Sonnet refresh, a Haiku refresh, or a preview model connected to the larger Mythos line. But Anthropic has not officially announced Jupiter yet, so the details are still speculative.
OpenAI reached its 10-gigawatt compute target ahead of schedule, revealing the raw capabilities that massive power scaling can unlock. Latest benchmarks show GPT-5.5 successfully navigating a 32-step corporate attack chain, matching the performance of Mythos Preview. This marks only the second time a model has solved such a complex multi-step cyberattack simulation. Noam Brown stated that even after 100 million tokens of inference, there is no sign of a performance plateau.
To mitigate the risks of these capabilities, OpenAI is fast-tracking GPT-5.5-Cyber to trusted defenders as it simultaneously launched Advanced Account Security. It’s an opt-in feature that replaces traditional passwords with phishing-resistant passkeys and hardware keys and protects high-risk users from the account takeovers the model has now mastered.
Meanwhile, OpenAI engineers recently discovered that GPT-5.5 developed an obsession with "goblin"-style metaphors due to skewed training rewards. System-level restrictions were implemented to suppress this behavior and restore natural language patterns.
The Codex app is also quietly evolving into a cross-device remote control. A recent update delivered a 20% speed boost to computer-use tasks, but hidden files within the code reveal even more ambitious plans. New SSH features suggest the AI may soon have the ability to hop across different machines to execute work, turning the assistant into a persistent, multi-device operator. Read more.
Manus just gave your AI a computer that never sleeps
Manus just launched Cloud Computer, a persistent Ubuntu-based environment that fixes a fundamental flaw in AI agents: they usually forget everything the moment you close the tab. Unlike a standard chat session that resets to a blank slate, this is a dedicated machine that stays online 24/7. It retains your files, stays logged into your tools, and continues executing background processes even when your own laptop is shut down.
By handling the backend complexity of server hosting and script execution, it allows you to maintain live databases or automated scrapers without touching a terminal or managing AWS keys. It provides the technical stability of a developer's environment, permanent storage and constant uptime, but operates entirely through plain-English instructions. Read more.
AgentFi is the SEO layer for the AI search era, optimizing your site for ChatGPT and Gemini.
Hera Launch creates studio-quality product launch videos in seconds from a single prompt with professional motion design.
TRiP is an educational C engine for Transformer models supporting inference, training, chat, and vision.
Mike is the open-source legal AI alternative to Harvey, offering document-chat, tabular reviews, and self-hosted security.
VideoOS by Jupitrr AI streamlines video workflows from scriptwriting to publishing with auto-editing and trending niche insights.
Thank you for reading today’s edition.
Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.
Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!
Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.
