Discussion | Link

I flew in a one-of-a-kind electric airplane and got a glimpse of what the future of aviation could look like. Taylor Rains/Business Insider Beta Technologies aims to revolutionize travel with its electric aircraft, the Alia CX300. I flew on the plane during Beta's media day and thought it was smooth and quiet. There are still questions around certification, infrastructure, battery life, and public acceptance. The future of aviation isn't a sleek jet — it's a tiny whale-shaped airplane that could turn The Jetsons fantasy of aerial commuting into reality. Vermont-based startup Beta Technologies aims to convince US consumers that electric aviation has arrived in ways that will change how they travel for work and leisure, and invited media to ride in the state-of-the-art all-electric aircraft that it calls the Alia CX300. For 20 minutes, we zipped above Burlington at over 100 miles per hour as test pilot Chris "Pooter" Caputo showed us how it banks, glides, and manages energy. The ride was smooth, quiet, and surprisingly fast, with sweeping views of Lake Champlain and the Green Mountains below. This new industry believes electric aircraft like the Alia CX300 will take over short regional routes from today's small fuel-powered turboprops and helicopters — making travel cleaner, cheaper, and quieter. Beta's CX300, a type of aircraft designated cTOL, already has orders from carriers like Air New Zealand and is expected to begin revenue-cargo flights later this year under a Transportation Department pilot program. Full certification is expected in late 2027. Beta is also developing an electric vertical takeoff and landing aircraft, or eVTOL — the Alia 250 — to certify in the years after its CX300. It mirrors much of the CX300's technology but takes off from a vertiport rather than a runway, making it particularly suitable for city commutes. This approach differs from that of other US competitors, such as Joby Aviation and Archer Aviation, which are primarily focused on eVTOL aircraft instead of stepped certification. Beta president and CEO Kyle Clark said it's a pragmatic strategy that ultimately makes vertical certification easier: "By the time you get the cTOL certified, you effectively have 80% of the requirements for the eVTOL." Still, the company faces major hurdles, particularly around charging infrastructure, certification, and public acceptance — especially if expected cost savings don't translate into affordable fares. And the up to $13 million in projected battery replacement sales over each aircraft's lifetime, outlined in a 2025 SEC filing, could spook potential customers. Here's a closer look at the Alia CX300 and what my flight was like. Alia CX300 can carry five passengers and one pilot. Taylor Rains/Business Insider The Alia CX300 can carry five passengers and one pilot, with no divider between the cockpit and the cabin. Inside, two rows of seats sit behind the pilot. My colleague Dan Allen, who sat in the back row at over six feet tall, said he had plenty of legroom. I sat in the front and felt the same. Flying from the cockpit gave me a bird's-eye view, and the visibility below seemed to eliminate some blind spots. That said, you do feel every bank and bump in a small aircraft like this, and some passengers may get motion sickness. Dramamine probably wouldn't hurt. It runs on rechargeable battery packs. The charging cubes are left of the aircraft. One unit charges the batteries, while the other cools them during the process. Taylor Rains/Business Insider Up to five battery packs stored in the belly of the aircraft provide around 250 kWh of energy — roughly 390 miles of range under ideal conditions. The redundancy in case a battery fails is crucial for safety and certification. While not fully certified yet, the company secured special federal permissions to operate demonstration flights like this one. Beta has spent years developing its electric propulsion system in-house, with separate versions for its cTOL and upcoming eVTOL aircraft. The batteries can be recharged in about an hour using large, cube-shaped charging units, which Beta also builds and sells to other operators for hundreds of thousands of dollars. The charging business is part of its broader strategy to diversify revenue beyond aircraft sales. By designing and building much of its technology in-house, Beta has greater control over quality and costs, Clark said — a vertical integration strategy similar to SpaceX, which also builds much of its rockets and supporting infrastructure itself. Our pilot cut the engines halfway through the flight. Taylor Rains/Business Insider To demonstrate the CX300's glide, Caputo cut the engines at about 1,300 feet. While it was slightly rattling, the aircraft maintained a controlled glide and gradually lost altitude. Caputo said that even in the event of complete power loss away from an airport, the aircraft could glide down onto a suitable landing strip. The engine itself is also built with redundancy. The rear-mounted electric propulsion unit is designed with independent power paths, meaning one can fail while the other continues operating. Our flight cost a few dollars. That doesn't mean fares will be cheap. BETA is also making an electric aircraft for the US military. Taylor Rains/Business Insider Our flight technically cost only a few dollars in electricity, Caputo said. Transportation Secretary Sean Duffy similarly took a demo flight with Beta last week that cost $3, per the company. A similarly sized fuel-powered Cessna would cost a few hundred dollars. But that doesn't necessarily translate into cheap fares. Operating costs still include pilots, maintenance, insurance, and infrastructure — so the true economics are still a big question. But electricity is still cheaper than jet fuel, especially as oil prices remain sky high amid the US war in Iran. Alia CX300 has already been tested on real-life missions. Caputo performed deep banks to show us the cTOL's maneuverability. Taylor Rains/Business Insider Beta has partnered with several airlines, including Republic Airways, Air New Zealand, and the UK's Loganair, to conduct real-world test flights. For example, Beta and ANZ completed more than 100 flights — including organ deliveries — across 12 airports and roughly 7,000 miles earlier this year. In the US alone, there are more than 4,000 public-use airports that are too rural or small for larger commercial jets, yet still need more affordable connections to the broader air network — think leisure places like the Hamptons and Catalina Island, or remote Alaskan communities that rely on regular supply flights. That network could become the backbone for the Alia CX300, flying short-haul routes within airline fleets or as a standalone service opening untapped routes at lower cost — ultimately laying the groundwork for Beta's vision of eVTOL flying taxis. "We're ready to go today with cargo cTOL aircraft," Clark said. "Tomorrow it's going to be passenger cTOL aircraft, then cargo VTOL aircraft, then passenger VTOL aircraft." Read the original article on Business Insider

Elizabeth Dwoskin / Washington Post: Inside the Trump admin's push to integrate AI into the healthcare system, including an FDA regulatory fast track for digital health tech like AI chatbots  —  The administration is laying the groundwork for chatbots that can diagnose illness and prescribe medicine, but physicians say AI can introduce more problems.

Ramsay Hodgson / Financial Times: Paris-listed Teleperformance, the world's largest customer service company, has become one of Europe's most shorted stocks, as hedge funds bet on AI disruption  —  Outsourcing companies hit as investors see ‘clean’ disruption risk  —  Hedge funds are betting against the shares and debt …

Abhirup Roy / Reuters: Sources: Uber has committed nearly $500M to self-driving startup Nuro, providing a crucial runway as Nuro works to prove its technology at commercial scale  —  Uber (UBER.N) has committed close to half a billion dollars in self-driving startup Nuro, two sources directly aware of the matter told Reuters …

Ami Yamada / Nikkei Asia: Kepple: seed-stage startup funding in Japan fell 42% YoY in 2025 to a 10-year low of $124M, as the Tokyo Stock Exchange moves to reduce small listings  —  TOKYO — Funding in Japan for the earliest-stage startups tumbled 42% last year to a 10-year low of 19.9 billion yen ($124 million) …

Mayumi Negishi / Bloomberg: SoftBank's PayPay, Japan's dominant payments app, says it will take a 70.2% stake in T&D Financial Life Insurance for $840M, expected to close in October 2027  —  SoftBank Group Corp.'s payments unit is buying the life insurance unit of T&D Holdings Inc. for ¥134.3 billion ($840 million) …

<p><strong>Release:</strong> <a href="https://github.com/simonw/micropython-wasm/releases/tag/0.1a2">micropython-wasm 0.1a2</a></p> <p>I added a CLI to <code>micropython-wasm</code> (<a href="https://github.com/simonw/micropython-wasm/issues/7">issue #7</a>), inspired by the first draft of <a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/">the blog entry</a> when I realized it would be a great way to illustrate the <a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#try-it-yourself">Try it yourself</a> section.</p> <p>Tags: <a href="https://simonwillison.net/tags/python">python</a>, <a href="https://simonwillison.net/tags/sandboxing">sandboxing</a>, <a href="https://simonwillison.net/tags/webassembly">webassembly</a>, <a href="https://simonwillison.net/tags/micropython">micropython</a></p>

<p>I've been experimenting with different approaches to running code in a sandbox for several years now, but my latest attempt feels like it might finally have all of the characteristics I've been looking for. I've released it as an alpha package called <a href="https://github.com/simonw/micropython-wasm">micropython-wasm</a>, and I'm using it for a code execution sandbox plugin for <a href="https://github.com/datasette/datasette-agent">Datasette Agent</a> called <a href="https://github.com/datasette/datasette-agent-micropython">datasette-agent-micropython</a>.</p> <ul> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#why-do-i-want-a-sandbox-">Why do I want a sandbox?</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#what-i-want-from-a-sandbox">What I want from a sandbox</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#webassembly-looks-really-promising-here">WebAssembly looks really promising here</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#micropython-in-webassembly">MicroPython in WebAssembly</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#building-the-first-version">Building the first version</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#try-it-yourself">Try it yourself</a></li> <li><a href="https://simonwillison.net/2026/Jun/6/micropython-in-a-sandbox/#should-you-trust-my-vibe-coded-sandbox-">Should you trust my vibe-coded sandbox?</a></li> </ul> <h4 id="why-do-i-want-a-sandbox-">Why do I want a sandbox?</h4> <p>My key open source projects - <a href="https://datasette.io/">Datasette</a>, <a href="https://llm.datasette.io/">LLM</a>, even <a href="https://sqlite-utils.datasette.io/">sqlite-utils</a> - all support plugins.</p> <p>I absolutely love plugins as a mechanism for extending software. A carefully designed plugin system reduces the risk involved in trying new things to almost nothing - even the wildest ideas won't leave a lasting influence on the core application itself. My software can grow a new feature overnight and I don't even have to review a pull request!</p> <p>There's one major drawback: my plugin systems all use Python and <a href="https://pluggy.readthedocs.io/en/latest/">Pluggy</a>, and plugin code executes with full privileges within my applications. A buggy or malicious plugin could break everything or leak private data.</p> <p>I'd love to be able to run plugin-style code in an environment where it is unable to read unapproved files, connect to a network, or generally operate in a way that's risky or harmful to the rest of the application or the user's computer.</p> <p>My interest covers more than just plugins. For Datasette in particular there are many features I'd like to support where arbitrary code execution would be useful. I've already experimented with this for <a href="https://enrichments.datasette.io/">Datasette Enrichments</a>, where code can be used to transform values stored in a table. I'd love to build a mechanism where you can run code on a schedule that fetches JSON from an approved location, runs a tiny bit of code to reformat it into a list of dictionaries, then inserts those as rows in a SQLite database table.</p> <h4 id="what-i-want-from-a-sandbox">What I want from a sandbox</h4> <p>My goal is to execute code safely within my own Python applications. Here's what I need:</p> <ul> <li>Dependencies that <strong>cleanly install from PyPI</strong>, including binary wheels across multiple platforms if necessary. I don't want people using my software to have to take any extra steps beyond directly installing my Python package.</li> <li>Executed code must be subject to both <strong>memory</strong> and <strong>CPU</strong> limits. I don't want <code>while True: s += "longer string"</code> to crash my application or the user's computer.</li> <li> <strong>File access must be strictly controlled</strong>. Either no filesystem access at all or I get to define exactly which files can be read and which files can be written to.</li> <li> <strong>Network access is controlled as well</strong>. Sandboxed code should not be able to communicate with anything without going through a layer I fully control.</li> <li>Support for interaction with <strong>host functions</strong>. A sandbox isn't much use if I can't carefully expose selected platform features to the code that it's running.</li> <li>It has to be <strong>robust, supported, and clearly documented</strong>. I've lost count of the number of sandbox projects I've seen in repos with warnings that they aren't actively maintained!</li> </ul> <h4 id="webassembly-looks-really-promising-here">WebAssembly looks really promising here</h4> <p>Web browsers operate in the most hostile environment imaginable when it comes to malicious code. Their job is to download <em>and execute</em> untrusted code from the web on almost every page load.</p> <p>Given this, JavaScript engines should be excellent candidates for sandboxes. Sadly those engines are also extremely complicated, and are not designed for easy embedding in other projects. Most of the V8-in-Python projects I've seen are infrequently maintained and come with warnings not to use them with completely untrusted code.</p> <p>WebAssembly is a <em>much better</em> candidate. It was designed from the start to support all of the characteristics I care about and has been tested in browsers for nearly a decade. The <a href="https://pypi.org/project/wasmtime">wasmtime</a> Python library brings WASM to Python, is actively maintained, and has binary wheels.</p> <h4 id="micropython-in-webassembly">MicroPython in WebAssembly</h4> <p>WebAssembly engines like wasmtime run WebAssembly binaries. Some programming languages like Rust are easy to compile directly to WebAssembly. Dynamic languages like JavaScript and Python are harder - they support language primitives like <code>eval()</code>, which means they need a full interpreter available at runtime.</p> <p>To run Python we need a full Python interpreter compiled to WebAssembly, wired up in a way that makes it easy to feed it code, hook up host functions and access the results.</p> <p>Pyodide offers an outstanding package for running Python using WebAssembly in the browser, but using Pyodide in server-side Python isn't supported. The most recent advice I could find was <a href="https://github.com/pyodide/pyodide/discussions/5145">from October 2024</a> stating "Pyodide is built by the Emscripten toolchain and can only run in a browser or Node.js".</p> <p>The other day I decided to take a look at <a href="https://micropython.org">MicroPython</a> as an option for this. The MicroPython site says:</p> <blockquote> <p>MicroPython is a lean and efficient implementation of the Python 3 programming language that includes a small subset of the Python standard library and is optimised to run on microcontrollers and in constrained environments.</p> </blockquote> <p>WebAssembly sure feels like a constrained environment to me!</p> <h4 id="building-the-first-version">Building the first version</h4> <p>I had GPT-5.5 Pro <a href="https://chatgpt.com/share/6a1e2a5c-58b8-8328-ba1c-0e6aadb0a051">do some research for me</a>, which turned up <a href="https://github.com/micropython/micropython/pull/13676">this PR against MicroPython</a> by <a href="https://github.com/yamt">Yamamoto Takahashi</a> titled "Experimental WASI support for ports/unix".</p> <p>It then produced this <a href="https://github.com/simonw/micropython-wasm/blob/c08fbd2276b15dc8c9bdff82845f750971f45647/research.md">research.md document</a>, so I let Codex Desktop and GPT-5.5 high <a href="https://gist.github.com/simonw/27461a16d76f28f8619c609444d544fe">loose on it</a> to see what would happen:</p> <blockquote> <p><code>read the research.md document and build this. You will probably need to write a script that compiles a custom WASM version of MicroPython as part of this project - fetch the MicroPython code to a /tmp directory for this as part of that script.</code></p> </blockquote> <p>It worked. I now had a prototype Python library that could execute Python code inside a WebAssembly sandbox!</p> <p>The trickiest piece to solve was persistent interpreter state. The WASM build we are using here exposes a single entry point which starts the interpreter, runs the code and then stops the interpreter at the end.</p> <p>This works fine for one-off scripts, but for Datasette Agent I want variables and functions to stay resident in memory so I can reuse them across multiple code execution calls.</p> <p>A neat thing about working with coding agents is that you can get from an idea to a proof of concept quickly. I prompted:</p> <blockquote> <p><code>For keeping variables resident: what if we ran code inside micropython itself which called a host function get_next_python_code() and then passed that to eval() - and that host function blocked until new code was available, maybe by running in a thread with a queue? Could that or a similar idea help here?</code></p> </blockquote> <p>After some iteration we got to a version of this that works! In Python code you can now do this:</p> <pre><span class="pl-k">from</span> <span class="pl-s1">micropython_wasm</span> <span class="pl-k">import</span> <span class="pl-v">MicroPythonSession</span> <span class="pl-k">with</span> <span class="pl-en">MicroPythonSession</span>() <span class="pl-k">as</span> <span class="pl-s1">session</span>: <span class="pl-en">print</span>(<span class="pl-s1">session</span>.<span class="pl-c1">run</span>(<span class="pl-s">"x = 10<span class="pl-cce">\n</span>print(x)"</span>).<span class="pl-c1">stdout</span>) <span class="pl-en">print</span>(<span class="pl-s1">session</span>.<span class="pl-c1">run</span>(<span class="pl-s">"x += 5<span class="pl-cce">\n</span>print(x)"</span>).<span class="pl-c1">stdout</span>) <span class="pl-en">print</span>(<span class="pl-s1">session</span>.<span class="pl-c1">run</span>(<span class="pl-s">"print(x * 2)"</span>).<span class="pl-c1">stdout</span>)</pre> <p>Under the hood this starts a thread, sets up a request queue and then sends messages to that queue for the <code>session.run()</code> command, each time waiting on a reply queue for the result of that execution. Inside WASM the MicroPython interpreter blocks waiting for a <code>__session_next__()</code> host function to return the next line of code, which it runs <code>eval()</code> on before calling <code>__session_result__({"id": request_id, "ok": True})</code> when each block has been successfully executed.</p> <p>The other piece of complexity was supporting host functions, so my Python library could selectively expose functions that could then be called by code running in MicroPython.</p> <p>Codex ended up solving this with <a href="https://github.com/simonw/micropython-wasm/blob/0.1a1/micropython_wasm/usercmodule/host/hostmodule.c">78 lines of C</a>, which ends up compiled into the <a href="https://github.com/simonw/micropython-wasm/blob/0.1a1/micropython_wasm/artifacts/micropython-wasi.wasm">362KB WebAssembly blob</a> I'm distributing with the package.</p> <p>I am by no means a C programmer, but I've read the C and had two different models explain it to me (here's <a href="https://claude.ai/share/62f74371-cc3c-44f2-b406-33d03513de9e">Claude's explanation</a>) and I've subjected it to a barrage of tests.</p> <p>The great thing about working with WebAssembly is that if the C turns out to be fatally flawed the worst that can happen is the WebAssembly execution will fail with an exception. I can live with that risk.</p> <p>Memory limits are directly supported by wasmtime. CPU limits are a little harder: wasmtime offers a "fuel" concept to limit how many operations a WebAssembly call can execute, and that's the correct fit for this problem, but the units are hard to reason about. I'm experimenting with a 20 million default "fuel" setting now but I'm not confident that it's the most appropriate value.</p> <h4 id="try-it-yourself">Try it yourself</h4> <p>The <code>micropython-wasm</code> alpha is now <a href="https://pypi.org/project/micropython-wasm">live on PyPI</a>.</p> <p>You can try it from your own Python code as <a href="https://github.com/simonw/micropython-wasm">described in the README</a>. I've also added a simple CLI mode in <a href="https://github.com/simonw/micropython-wasm/releases/tag/0.1a2">version 0.1a2</a> which means you can try it using <code>uvx</code> without first installing it like so:</p> <div class="highlight highlight-source-shell"><pre>uvx micropython-wasm -c <span class="pl-s"><span class="pl-pds">'</span>print("Hello world")<span class="pl-pds">'</span></span> <span class="pl-c"><span class="pl-c">#</span> To see it run out of fuel:</span> uvx micropython-wasm -c <span class="pl-s"><span class="pl-pds">'</span>s = ""; while True: s += "longer"<span class="pl-pds">'</span></span> <span class="pl-c"><span class="pl-c">#</span> Outputs: micropython-wasm: guest exited with code 1</span></pre></div> <p>You can also try it in <a href="https://agent.datasette.io/">Datasette Agent</a> like this:</p> <div class="highlight highlight-source-shell"><pre>uvx llm keys <span class="pl-c1">set</span> openai <span class="pl-c"><span class="pl-c">#</span> Paste in an OpenAI key, then:</span> uvx --with datasette-agent \ --with datasette-agent-micropython \ --prerelease allow \ datasette --internal internal.db \ -s plugins.datasette-llm.default_model gpt-5.5 \ --root -o</pre></div> <p>Then navigate to <a href="http://127.0.0.1:8001/-/agent">http://127.0.0.1:8001/-/agent</a> and run the prompt:</p> <blockquote> <p><code>show me some micropython</code></p> </blockquote> <p><img src="https://static.simonwillison.net/static/2026/micropython-in-datasette-agent.jpg" style="max-width: 100%" alt="Screenshot of a chat application interface with a dark blue-grey header reading &quot;home&quot; on the left and &quot;root&quot; with a hamburger menu icon on the right. Below is a navigation row with &quot;← Back&quot; and &quot;Chat&quot; on the left and an &quot;EXPORT&quot; button on the right. A blue user message bubble reads &quot;show me some micropython&quot;. Below it a collapsed thinking section reads &quot;▸Thinking: … to show the result clearly. After that, I can wrap up with a brief explanation!&quot; followed by a &quot;▶ Tool: execute_micropython&quot; label. A code block follows: &quot;# A tiny MicroPython example: blink-style logic + Fibonacci&quot; / &quot;def fib(n):&quot; / &quot; a, b = 0, 1&quot; / &quot; out = []&quot; / &quot; for _ in range(n):&quot; / &quot; out.append(a)&quot; / &quot; a, b = b, a + b&quot; / &quot; return out&quot; / &#39;print(&quot;Hello from MicroPython!&quot;)&#39; / &#39;print(&quot;First 10 Fibonacci numbers:&quot;, fib(10))&#39; / &quot;# MicroPython often runs on microcontrollers, e.g.:&quot; / &quot;# from machine import Pin&quot; / &quot;# led = Pin(2, Pin.OUT)&quot; / &quot;# led.value(1) # turn LED on&quot; / &quot;# led.value(0) # turn LED off&quot;. Below a horizontal divider is the output: &quot;Hello from MicroPython!&quot; / &quot;First 10 Fibonacci numbers: [0, 1, 1, 2, 3, 5, 8, 13, 21, 34]&quot;, followed by a &quot;▶ Result: execute_micropython&quot; label. At the bottom is a text input field with placeholder &quot;Type a message...&quot; and a blue &quot;Send&quot; button." /></p> <p>You can try a live demo of that plugin running in Datasette Agent by signing into <a href="https://agent.datasette.io">agent.datasette.io</a> with your GitHub account.</p> <h4 id="should-you-trust-my-vibe-coded-sandbox-">Should you trust my vibe-coded sandbox?</h4> <p>Having complained about immature, loosely-maintained sandboxing libraries, it's deeply ironic that I've now built my own!</p> <p>I deliberately slapped an alpha release version on it, and I'm not ready to recommend it to anyone who isn't willing to take a significant risk.</p> <p>I've put it through enough testing that I'm OK using it myself. I've shipped my first plugin that uses it, <a href="https://github.com/datasette/datasette-agent-micropython">datasette-agent-micropython</a>. I've also locked GPT-5.5 xhigh in that Datasette Agent plugin and <a href="https://gist.github.com/simonw/5de497c44d25f9fd459c8aa2c959fe4a">challenged it to break out of the sandbox</a> and so far it has not managed to.</p> <p>I'm hoping this implementation can convince some companies with professional security teams and high-stakes problems to commit to using Python in WebAssembly as a sandboxing approach and open source their own solutions.</p> <p>Tags: <a href="https://simonwillison.net/tags/python">python</a>, <a href="https://simonwillison.net/tags/sandboxing">sandboxing</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/webassembly">webassembly</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/ai-assisted-programming">ai-assisted-programming</a>, <a href="https://simonwillison.net/tags/codex">codex</a>, <a href="https://simonwillison.net/tags/datasette-agent">datasette-agent</a>, <a href="https://simonwillison.net/tags/micropython">micropython</a></p>

Igor Bonifacic / Engadget: OpenAI rolls out Lockdown Mode, an optional security setting designed to offer users advanced protection from prompt injection attacks by limiting some features  —  The company says most users don't need to use the feature.  —  OpenAI has begun rolling out Lockdown Mode …

Discussion | Link

Kate Conger / New York Times: Google Cloud says its SpaceX compute deal is a “short-term” agreement “to ensure we have bridge capacity to meet surging customer demand” for Gemini Enterprise  —  Elon Musk's rocket company said Google would pay it $920 million a month, as it prepared for its initial public offering.

Akash Girimath / Decrypt: Privacy token Zcash plunges after the disclosure of a 2022 vulnerability in its Orchard shielded pool that could have allowed undetectable ZEC counterfeiting  —  Zcash plunged double digits overnight after developers disclosed a critical vulnerability in the protocol's Orchard shielded pool …

<p><strong><a href="https://help.openai.com/en/articles/20001061-lockdown-mode">OpenAI Help: Lockdown Mode</a></strong></p> OpenAI first teased this <a href="https://openai.com/index/introducing-lockdown-mode-and-elevated-risk-labels-in-chatgpt/">in February</a>, but now it's live and "rolling out to eligible personal accounts, including Free, Go, Plus, and Pro, and self-serve ChatGPT Business accounts":</p> <blockquote> <p>Lockdown Mode is designed to help prevent the final stage of data exfiltration from a prompt injection attack by limiting outbound network requests that could transfer sensitive data to an attacker. Lockdown Mode does not prevent prompt injections from appearing in the content ChatGPT processes. For example, a prompt injection could appear in cached web content or in an uploaded file, and could still affect the behavior or accuracy of a response.</p> </blockquote> <p>This looks really good to me.</p> <p>The <a href="https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/">Lethal Trifecta</a> occurs when an LLM system has access to all three of access to private data, exposure to untrusted content and a way to steal data and transmit it back to the attacker.</p> <p>The only way to solve the trifecta is to cut off one of the three legs, and by far the easiest leg to restrict without making your LLM systems far less useful is the exfiltration vectors to steal data.</p> <p>It looks to me like lockdown mode directly attacks that leg, using mechanisms that are deterministic and, crucially, are not evaluated by AI systems that themselves can be subverted by sufficiently devious attacks.</p> <p>The existence of lockdown mode does however imply that ChatGPT, in its default settings, does <em>not</em> provide robust protection against sufficiently determined data exfiltration attacks!</p> <p><strong>Update</strong>: <a href="https://twitter.com/cryps1s/status/2062923575049531422">This tweet</a> OpenAI CISO Dane Stuckey:</p> <blockquote> <p>Lockdown mode is not meant for everyone. However, for folks who have an elevated risk profile - due to who they are, what they work on, or the types of data they work with - it's an excellent tool for further securing themselves. This has some tradeoffs on functionality and utility, but for these users, the tradeoff is worthwhile.</p> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/security">security</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/openai">openai</a>, <a href="https://simonwillison.net/tags/prompt-injection">prompt-injection</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/lethal-trifecta">lethal-trifecta</a></p>

David Jeans / Reuters: Sources and docs detail defense tech startup Shield AI's struggles to overcome years of technical hitches and safety concerns with its V-BAT autonomous drone  —  A year ago, Ryan Tseng, the head of U.S. defense tech startup Shield AI, announced his company had turned a new page.

The 2026 World Cup is shaping up to be the most profitable tournament for FIFA and the most expensive for fans in history. Business Insider reporter Pete Syme investigated why ticket prices are so high and compiled five charts that illustrate just how out of reach many of the tournament's games are for average fans. What Syme discovered is that ticket prices are higher than ever because FIFA has adopted ticketing practices common in North America, like dynamic pricing and running its own secondary resale market, where prices skyrocket further. But it doesn't stop there. The additional costs fans have to pay are also high, including flights, hotels, and transit to game stadiums. Some American cities are jacking up transit prices on game days. When it's all said and done, FIFA stands to make $13 billion during the 2026 World Cup cycle. Read the original article on Business Insider

In the end, the three companies involved all point the finger at each other.

After a very profitable decade on Microsoft's board, Reid Hoffman is stepping down to focus on his AI drug discovery startup Manus.

Antonio Gracias has served on SpaceX's board since 2010. Bloomberg/Getty Images SpaceX's S-1 shows that entities controlled by Antonio Gracias own roughly 7.3% of Class A stock ahead of the IPO. Based on the $1.5 trillion valuation investors expect SpaceX to achieve, that stake would be worth roughly $90 billion. That level of wealth could instantly vault Gracias to the ranks of some of the richest people in the world. Antonio Gracias, one of Elon Musk's closest confidants and an early SpaceX investor, could land a massive multibillion-dollar windfall from the company's upcoming IPO. The SpaceX S-1 filing shows that Gracias owns more than 500 million shares of SpaceX stock through investment firms affiliated with Valor Equity Partners, the private equity firm he founded. The filing says those holdings represent roughly 7.3% of SpaceX's Class A stock ahead of the IPO. Based on a $1.5 trillion valuation some investors expect SpaceX to achieve in its IPO, those holdings would be worth roughly $91.6 billion. The amount would mark one of the biggest fortunes ever created by a venture investor. Gracias' relationship with Musk dates back more than two decades. The two met through Silicon Valley and PayPal circles in the early 2000s, shortly after Musk sold PayPal to eBay. At the time, Gracias was growing Valor Equity Partners, the Chicago-based investment firm he founded in 1995. Gracias became one of Musk's most trusted confidants and one of the small group of executives Musk repeatedly leaned on for advice. When Tesla faced repeated financial challenges during the late 2000s, and again during the Model 3 production ramp, Gracias emerged as one of Musk's strongest defenders inside Tesla's boardroom. He was also involved across multiple other Musk companies as the billionaire entrepreneur expanded into rockets, AI, satellites, and infrastructure with SpaceX and xAI. Valor became one of the earliest major institutional investors in SpaceX and steadily expanded its position while the company remained private for more than two decades. The SpaceX filing shows the relationship now extends well beyond stock ownership. According to the S-1 filing, subsidiaries tied to xAI entered into nearly $20 billion in lease agreements with Valor-affiliated entities for AI infrastructure and computing equipment. Those contracts generated hundreds of millions of dollars in payments to these Valor-related entities during 2025 and early 2026. Nosek and Shotwell will also make billions The filing shows how concentrated SpaceX stock ownership remains among the rest of Musk's longtime allies. Luke Nosek, a fellow PayPal alum and early Musk associate, owns nearly 33 million SpaceX shares through direct holdings and Nosek Capital. At the current IPO valuation, his stake would be worth roughly $6 billion. Nosek also co-founded VC firm Founders Fund. Gwynne Shotwell, SpaceX's president and chief operating officer, owns roughly 12.6 million shares of SpaceX. Her holdings would be worth about $2.3 billion at the estimated IPO valuation. Shotwell joined SpaceX in 2002 as employee No. 11 and became one of the company's most impactful executives, helping secure NASA contracts and overseeing the growth of Falcon launches and Starlink into businesses. Other directors own smaller but still substantial positions. Ira Ehrenpreis, the VC best known for serving on Tesla's board, owns about 1.37 million SpaceX shares. Those holdings would be worth roughly $250 million. Randy Glein, founder of VC firm DFJ Growth, owns about 278,000 shares. His holdings would be worth roughly $50 million. A new filing released Thursday indicated that the company aims to raise $75 billion in its initial public offering. It plans to offer 555,555,555 shares of common stock at $135 per share. That offering would be over twice the prior record for the largest IPO, set by Aramco when it raised $29 billion in its 2019 debut. Read the original article on Business Insider