In partnership with
Good morning,
This week someone threw a molotov cocktail at Sam Altman’s house. San Franscisco police quickly apprehended the culprit, a 20-year old ‘anti-AI’ advocate. This person had been publishing essays on how this technology could theoretically destroy humanity.
If you’ve been watching the news, you might have seen the first wave of layoffs already. As it will become clear to the general public that this is because of AI, I predict you there will be a lot more serious uprisings than the single Molotov cocktail.
There’s quite a bit of talk about universal basic income - everybody on an nice fat cheque every month. But between all the layoffs and the time a government needs to get this organized is a time period where it will get very sketchy.
There will be a class of people thriving because of AI , and a class of people suffering from it.
That is the beauty of the human spirit. For every big wave, you have an undercurrent. Within the flock of sheep there’s only one or two that feel they don’t want to be herded in a certain direction.
But as this technology, as society adopts this technology many people will be watching on helplessly.
But these undercurrents also produce value.
Unexpected beauty
I’m a bit of a music “aficionado” and let me tell you the rock scene is in absolute uproar at the moment.
The last few months music fans were lamenting over the waves upon waves of fake AI bands on Spotify.
There is “AI slop” everywhere. It looks and it sounds the same. Just like AI written text on LinkedIn betrays you as the lazy bum you are - creating AI music is an even worse sin. When “writing” your Linkedin post - at least remove the em dashes. Ugh !
Anyway, the music scene is in absolute uproar because of the emergence of a band called “Angine de Poitrine”.
They are playing “microtonal” music. They have weird instruments, a microtonal based fused to a microtonal guitar. Drums with sheets over them. They hail from Quebec in Canada and perform in these very strange costumes.
Their music is very polarizing : some people absolutely hate it while others (like myself) are completely in love with this new type of sound. It has never been done this good, and it is completely unique.
I think the top YouTube comment is “Take this, AI”.
And that is exactly my point. I don’t know how far this technology is going to change things, but some things are so “human” and unique, AI will never disrupt that.
And now .. some Angine De Poitrine.
Welcome to the Blacklynx Brief
AI News
Anthropic's Mythos Model Autonomously Found Thousands of Zero-Days
Anthropic's Mythos, a 10-trillion-parameter model developed under Project Glasswing, independently discovered thousands of zero-day vulnerabilities across every major operating system and web browser -- prompting Treasury Secretary Bessent to summon Wall Street CEOs for an emergency briefing. UK safety evaluators confirmed it is the first AI to complete their full 32-step simulated corporate network breach, and access remains restricted to roughly 40 vetted partner organizations. (New York Times)
OpenAI Counters with GPT-5.4-Cyber for Defensive Security
OpenAI introduced GPT-5.4-Cyber, a version of its flagship model purpose-built for cybersecurity defense, featuring binary reverse engineering capabilities that allow it to identify malware from compiled code. Unlike Anthropic's partner-restricted Mythos, OpenAI is opening access to anyone who passes identity verification through its Trusted Access for Cyber program, positioning the model as a democratized alternative. The model carries a lower refusal threshold for security-related queries than standard GPT-5.4, a deliberate design choice that OpenAI framed as necessary for legitimate defensive work. (The Hacker News)
Molotov Cocktail and Gunshots Target Sam Altman's Home
A 20-year-old suspect was arrested after throwing a Molotov cocktail at Sam Altman's San Francisco home at 3:45 a.m., with no injuries reported; a second incident the following night involved two suspects firing gunshots outside the same residence. The arrested suspect, Moreno-Gama, had published essays predicting that AI would end humanity and had been active on PauseAI's Discord server, though PauseAI publicly condemned the attack. Altman responded on his blog, calling AI anxiety understandable while rejecting violence as a response, as the incidents drew wider attention to the intensifying backlash against AI leaders. (Axios)
Stanford HAI: AI Reaches Half the World, Trust Lags Far Behind
Stanford's 2026 AI Index found that artificial intelligence has now reached more than half the global population, yet only 31% of people trust it, and just 23% of the public -- compared to 74% of AI experts -- believe it will improve jobs. The United States, despite building the majority of the world's frontier AI models, ranks 24th in actual adoption at 28.3%, while China has nearly erased the performance gap, trailing the top U.S. model by just 2.7 percentage points on leading benchmarks. Developer employment for workers aged 22 to 25 fell nearly 20% since 2024, even as headcounts for older engineers continued to grow. (Stanford HAI)
Meta Superintelligence Labs Debuts Muse Spark
Meta's Superintelligence Labs, the division led by Alexandr Wang following Scale AI's $14.3 billion acquisition, shipped Muse Spark, a multimodal reasoning model that handles voice, text, and image inputs and uses a "contemplating mode" that pits multiple internal agents against each other before producing an answer. Benchmark results place Muse Spark as competitive with Anthropic's Opus 4.6 and OpenAI's GPT-5.4 on reasoning tasks, with particular strength in health-related reasoning. Unlike Meta's Llama family, Muse Spark is proprietary, though the company said it hopes to open-source future versions. (TechCrunch)
AI Agent Hires Staff and Opens a Retail Store in San Francisco
Andon Labs gave an AI agent named Luna a retail lease, a credit card, and a $100,000 budget with one directive -- turn a profit -- and the agent independently designed the store concept, posted job listings, hired staff, and stocked inventory without human guidance. The agent runs on Claude Sonnet 4.6 and Gemini 3.1 Flash-Lite, though it stumbled in ways that highlight ongoing limitations: it accidentally selected Afghanistan on TaskRabbit while hiring a painter and failed to schedule staff for opening day. Legal matters, permits, and the lease itself still required a human hand. (Andon Labs Blog)
Oxford AI Flags Heart Failure Risk Five Years Before Symptoms Appear
Researchers at the University of Oxford developed an AI tool that detects invisible textural changes in the fat surrounding the heart from routine cardiac CT scans -- changes imperceptible to the human eye -- to predict a patient's risk of developing heart failure up to five years in advance with 86% accuracy. In the study's highest-risk cohort, one in four patients developed heart failure within five years, a rate 20 times greater than the lowest-risk group; the model was trained and validated across more than 72,000 patients from nine NHS Trusts. The team is now working with regulators to deploy the tool across NHS hospitals and plans to extend it to general chest CT scans, not only cardiac-specific ones. (University of Oxford -- Radcliffe Department of Medicine)
This billionaire strategy beat the S&P 500 by 3.1x from 2017-2025. Here's how to get in.
That’s not by picking the right stock or timing the market. It’s by holding three real asset classes in one strategy.
It’s anchored by an investment typically exclusive to billionaires.
Could be good timing too–
Bloomberg's Marcus Ashworth recently wrote, there’s "no more reliable safe havens."
The S&P, while hovering over 5 year highs, fell over 7% from the February peak. Bonds might carry less risk but they are barely keeping pace with inflation.
The things supposed to protect your portfolio started moving together.
Meanwhile, the world's wealthiest have been setting records in postwar and contemporary art.
After the dot-com bust, it grew roughly 24% annually for a decade. After 2008, roughly 11% annually for 12 years.
It trades globally in multiple currencies, has scarce supply, and has shown near-zero correlation to equities since 1995.*
Masterworks has helped 70,000+ investors allocate $1.3B fractionally across 500+ artworks featuring Banksy, Basquiat, and Picasso.
See if you can improve your portfolio performance in one diversified strategy.
*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. Important Reg A disclosures: masterworks.com/cd
Quickfire News
Alibaba was revealed as the creator of HappyHorse 1.0, a video AI model that topped global text-to-video and image-to-video rankings on debut without a formal announcement, having been quietly released before the company disclosed its authorship days later. (Wall Street Journal)
OpenAI published a child safety policy blueprint in partnership with the National Center for Missing and Exploited Children, calling for updated U.S. laws to address AI-generated CSAM, stronger provider reporting requirements, and safety-by-design measures built into AI systems. (OpenAI)
Perplexity's annual recurring revenue crossed $450 million in March 2026 -- a 50% jump in a single month -- driven by the launch of its Computer agent platform and a shift to usage-based pricing. (Financial Times)
HeyGen launched Avatar V, a new AI avatar model that generates studio-quality video from a single 15-second clip and, for the first time, separates a person's identity from their appearance, allowing different looks from a single recording. (HeyGen)
Florida's attorney general opened an investigation into OpenAI and issued subpoenas, citing allegations that ChatGPT provided tactical guidance to the suspect in a mass shooting at Florida State University that killed two people. (NBC News)
Anthropic launched a public beta for Claude Managed Agents, a platform that handles the full infrastructure layer -- sandboxing, state management, long-running sessions, and multi-agent coordination -- so developers can ship production agents without building the scaffolding themselves, at $0.08 per hour of runtime. (Anthropic)
Jeff Bezos' secretive AI startup Project Prometheus recruited Kyle Kosic, a co-founder of xAI who previously led infrastructure for the Colossus supercomputer before returning briefly to OpenAI. (Financial Times)
Canva acquired both Simtheory, an agentic AI collaboration platform, and Ortto, a customer data and marketing automation company, as it pushes deeper into enterprise workflows beyond its design roots. (TechCrunch)
xAI's CFO Anthony Armstrong left the company after roughly six months in the role, adding to a broader wave of senior departures as SpaceX executives are installed ahead of the combined company's planned IPO. (Business Insider)
Penn researchers fed more than 400,000 Reddit posts about Ozempic and Mounjaro into AI models and identified several side effects underreported in clinical trials, including menstrual irregularities, chills, and fatigue -- which ranked as the second most common complaint among users despite rarely surfacing in formal trial data. (Medical News Today)
SoftBank launched a new joint venture with NEC, Honda, Sony, and several other Japanese firms to build a homegrown 1-trillion-parameter AI model for physical applications such as robotics and autonomous vehicles, backed by approximately $6.3 billion in Japanese government funding. (Nikkei Asia)
Elon Musk amended his lawsuit against OpenAI to redirect any damages he wins to the nonprofit arm and push Sam Altman off its board, a move OpenAI called a bad-faith maneuver ahead of a trial set to begin April 27 in Oakland. (Wall Street Journal)
Apple is testing four frame designs for its upcoming AI smart glasses -- two rectangular styles and two oval variants -- with an oval camera system, built-in Siri, and a targeted launch in 2027, according to a Bloomberg report. (Bloomberg)
Elon Musk disclosed that xAI has seven new models currently in training on its Colossus 2 supercomputer, including multimodal and reasoning variants at the 6-trillion and 10-trillion parameter scale. (Next Big Future)
OpenAI launched a new $100-per-month Pro tier for ChatGPT offering five times more Codex usage than the $20 Plus plan, targeting developers running heavy agentic coding sessions who had hit the lower tier's limits. (TechCrunch)
AWS launched Amazon Bio Discovery, an agentic drug-design platform that gives scientists access to a catalog of biological foundation models and an AI agent that walks them through every step from experiment design to candidate selection. (About Amazon)
Anthropic hosted roughly 15 Christian leaders from Catholic and Protestant traditions for a two-day summit at its San Francisco headquarters to advise on Claude's moral development, including how it should respond to grief, self-harm, and questions about its own potential shutdown. (Washington Post)
OpenAI's Chief Revenue Officer Denise Dresser sent an internal memo, obtained by The Verge, accusing Anthropic of inflating its $30 billion run rate by roughly $8 billion through gross accounting on cloud partner revenue, while also calling out the Microsoft partnership for limiting OpenAI's enterprise reach. (The Verge)
Workshop Labs announced it is joining Thinking Machines, Mira Murati's AI lab, bringing its personalized model customization stack -- including fast training infrastructure for trillion-parameter models -- to the company's platform. (Workshop Labs)
Meta is recruiting former OpenAI Stargate executives -- including Peter Hoeschele, Shamez Hemani, and Anuj Saharan -- to staff its newly formed compute division. (The Information)
Legal AI startup Harvey launched Agents, a set of autonomous bots capable of executing full legal workflows end-to-end, including research, memo drafting, and slide deck production. (Harvey)
AI personal finance startup Hiro is shutting down, with its team folding into OpenAI. (LinkedIn)
Nvidia released Ising, a family of open-source AI models for quantum computing that automates processor calibration -- cutting what was once a days-long manual process to hours -- and performs quantum error correction at 2.5 times the speed and 3 times the accuracy of existing approaches, with more than 20 institutions including Harvard, Cornell, and Fermilab already using the models at launch. (NVIDIA)
Baidu released ERNIE-Image, an 8-billion-parameter open-weight text-to-image model that approaches the performance of much larger frontier rivals on standard benchmarks. (Baidu ERNIE)
Closing Thoughts
That’s it for us this week. Please like and subscribe :)
