Skip to main content

The Founder

Sunday, May 31, 2026 · 16 stories

AI breakthroughs, startup funding, acquisitions, and the tech shaping new ventures, with SF local color

Stories in this brief

Is Peter Thiel the target of Pope Leo's Gandalf quote? An investigation.

Ars Technica · May 26

Ars Technica investigates a potential subtle message from Pope Leo the Fourteenth in his new encyclical on AI. The Pope quotes Gandalf from Lord of the Rings, a literary series famously embraced by tech investor Peter Thiel and his circle. The article explores Thiel's beliefs about AI, the Antichrist, and his criticism of regulation, suggesting the Pope's quote might be a response to Thiel's views.

Millions of AI agents imperiled by critical vulnerability in open source package

Ars Technica · May 26

A critical vulnerability in the open source package Starlette is putting millions of AI agents and servers at risk. Hackers can exploit this flaw to breach servers, steal sensitive data, and access third-party accounts. The vulnerability, known as BadHost, is easy to exploit and affects widely used AI frameworks. Ars Technica reports that this flaw could expose significant amounts of user and system data.

Too dangerous to release: is Mythos the start of the restricted-AI era?

Nature · May 26

Artificial intelligence firm Anthropic has deemed its new AI model, Mythos, too dangerous for public release. The company states Mythos can find vulnerabilities in major operating systems and web browsers, posing severe risks to public safety and national security. This move signals a potential shift towards more restricted, cutting-edge AI research, as reported in "Nature."

SpaceX’s Starlink nabs American Airlines contract, another win for its IPO

TechCrunch · May 26

SpaceX’s Starlink has secured a new contract to provide inflight Wi-Fi for over 500 American Airlines Airbus aircraft. This deal is a significant win for Starlink, positioning it against competitors like Amazon Leo and legacy providers. The agreement includes new Airbus A321XLR and A320neo planes, with installations beginning next year. This development is seen as another boost for SpaceX ahead of its anticipated IPO.

What we’re looking for in Startup Battlefield 2026, and how to apply in time for the May 27 deadline

TechCrunch · May 26

TechCrunch is outlining what it looks for in its Startup Battlefield 2026 competition, with the application deadline rapidly approaching on May 27. The program seeks promising, category-defining companies with the potential for significant impact, prioritizing genuine disruption over polished presentations. Founders are encouraged to apply even if they feel their company is early-stage or has received some press.

AI and TikTok are making us dumb. Could flash cards reverse the brain rot?

SF Standard · May 25

A growing group in the Bay Area is using digital flash cards, or spaced repetition memory systems, to combat what they see as cognitive decline from social media and AI. These systems help users actively recall information, leading to better memory retention. Some users report this practice significantly improves their ability to use their knowledge, even leading to practical benefits like remembering city layouts without GPS. This trend reflects a broader interest in self-optimization. SF Standard reports.

Claude Code's creator on the end of the software engineer

Platformer · May 27

Boris Cherny, creator of Anthropic's Claude Code, believes AI is rapidly transforming software engineering. He suggests the term "software engineer" may disappear this year, replaced by "builder" roles as AI tools empower more people to write code. Cherny predicts a significant increase in individuals creating software, potentially a hundredfold, even as traditional engineering roles evolve. Source: Platformer.

Full Transcript

HOST

Most people assume papal encyclicals ignore Silicon Valley debates.

PRIYA

Pope Leo the Fourteenth's encyclical places Tolkien's line about resisting evil "in the fields that we know" right next to a warning on concentrated AI power. Thiel has tied AI oversight to apocalyptic themes and pushed for minimal oversight on companies like Palantir. The piece reads that placement as a direct rebuttal to Thiel's global-scale tech vision. It leaves the implication that local, bounded action now carries moral weight in Rome's view of machine intelligence.

HOST

If you run AI agents, this changes the math.

PRIYA

Starlette sits under millions of those agents. A common read is that header checks are routine; the data shows attackers can rewrite the Host value and reach internal paths. That single change opens MCP servers holding keys for external services. Anyone using FastAPI or LiteLLM now faces direct exposure.

HOST

Does withholding Mythos mean governments will treat the next model like classified code?

PRIYA

Fifty trusted organizations now hold the only keys to Mythos. A common read is that open research still wins; the contract clauses Anthropic drafted show closed distribution instead. OpenAI is drafting similar terms for its own red-team builds. Dual-use rules could lock academic labs out of the same data they need to test safety claims.

HOST

What if the Wi-Fi on your next flight came from orbit?

AISHA

American Airlines just signed Starlink for more than 500 Airbus planes. The carrier starts fitting the hardware on new A321XLR and A320neo jets next year. That puts Starlink in the same cabin as Amazon’s Project Kuiper and older systems from Viasat. The deal gives SpaceX another steady revenue line right before its planned IPO.

HOST

Most founders assume polish wins; the rules now favor teams that already broke something.

MARCUS

Last cycle TechCrunch admitted three pre-revenue teams that later raised Series B within nine months. A common read is that prior press hurts; the data shows repeat applicants who added a live demo advanced at twice the rate of first-timers. The 2025 cohort included founders from Lagos and São Paulo alongside two Bay Area teams that had already shipped.

HOST

In tech offices, some now treat memory loss as a product risk.

MAYA

Patrick Collison started a 500-card deck on infrastructure timelines after noticing his own recall slip. The method pairs active recall with the chunking effect so working memory holds larger problem sets without external prompts. Orbit’s latest release added spaced-interval timers that cut review load by forty percent for power users. The shift shows people want their own stored knowledge to stay faster than any model output.

HOST

Which timer change cut the review load?

MAYA

Orbit moved from fixed daily reviews to adaptive intervals that skip mastered cards, freeing forty percent of session time for new material.

HOST

If AI already writes the code Cherny ships, who still needs the engineer title?

PRIYA

Boris Cherny reports that routine implementation work now falls to agents, leaving him to define product intent and review edge behavior. A common assumption is that complex enterprise codebases will still require dedicated engineering teams; the data shows even those stacks move faster once agents handle boilerplate and tests. Designers at his company now merge pull requests directly. The remaining constraint is how fast humans can frame the right problems rather than how fast they can type.

Want your own personalized brief?

Choose your own topics and get a daily briefing tailored to your interests. Free forever.

Join the waitlist