Katie Tarasov / CNBC: Intel reports Q1 revenue up 7% YoY to $13.58B, vs. $12.42B est., and forecasts Q2 revenue and adjusted EPS above estimates; INTC jumps 15%+ after hours  —  - Intel reported first-quarter earnings that beat Wall Street expectations … Intel reported first-quarter earnings Thursday …

In March 2026, three LLM agents generated over 600,000 lines of code, ran 850 experiments, and helped secure a first-place finish in a Kaggle playground... In March 2026, three LLM agents generated over 600,000 lines of code, ran 850 experiments, and helped secure a first-place finish in a Kaggle playground competition. Success in modern machine learning competitions is increasingly defined by how quickly you can generate, test, and iterate on ideas. LLM agents, combined with GPU acceleration, dramatically compress this loop. Historically… Source

Monstera Production/Pexels The legendary American bank robber Willie Sutton spent 40 years robbing banks because, as he claimed in his autobiography, he loved doing it. And when asked why he chose banks of all places to rob, he allegedly replied “Because that’s where the money is.” Back in 2017, I wrote a book predicting it wasn’t just lovable rogues like Sutton who would soon be robbing banks, but artificial intelligence (AI). That day, it appears, could now be about to arrive. Banks around the world are seriously worried cyber criminals will soon take advantage of the latest advances in AI to try to rob them. The digital back door into the vault The finance world’s concern rests on the impressive cyber capabilities of a product called “Mythos”. This is the latest and most capable AI model from Anthropic, the company behind the popular Claude chatbot. As a member of the public, you can’t access or use this model – for now. That’s because Anthropic (and many others) believe Mythos is too capable to launch upon an unsuspecting world. Internal testing of Mythos has uncovered thousands of severe security vulnerabilities across every major operating system and web browser. Some of these vulnerabilities have gone undetected for decades. Many are what tech insiders call “zero day” vulnerabilities – attacks that are so dangerous that developers need to fix them in zero days’ time. Not for public use To counter this emerging threat, Anthropic has made the model available to a dozen partners of a defensive coalition that includes Microsoft, Amazon Web Services, Apple, Cisco and the Linux Foundation. The company has also committed US$100 million (about A$140 million) in usage credits and US$4 million (about A$5.6 million) in open-source grants to start finding and fixing these bugs. In addition, more than 40 additional organisations – including a number of US banks – have also received access. But worryingly, as far as we know, Anthropic has not yet granted access to any banks in Australia, the United Kingdom or Europe. To add to concerns, on Wednesday, Anthropic confirmed it was investigating claims in a Bloomberg report that a small group of unauthorised users had gained access to Mythos. However, at this stage, there is no suggestion this alleged access was for malicious purposes. Claude Mythos and Project Glasswing: why an AI superhacker has the tech world on alert Should you be worried? Last week, regulators and policymakers from around the world gathered at the International Monetary Fund spring meeting in Washington. The Iran war was a major focus. But attendees also issued a series of warnings about this emerging cybersecurity threat to the banking industry. Not only are banks an attractive target, being where the money is, but the industry runs on many legacy systems, decades old technology that may be especially vulnerable to these sorts of attacks. You personally don’t need to be too worried. Many countries have strong protections for bank customers. In Australia, for example, the first A$250,000 of a customer’s deposits are insured through the government-backed Financial Claims Scheme. And the Australian Securities and Investments Commission ensures banks investigate and reimburse fraudulent transactions where the customer is not at fault. So, it’s probably not a wise idea to withdraw your cash and put it under the mattress. But banks should be (and are) rushing to plug these vulnerabilities. I would recommend you regularly update your computer and smartphone to have the latest operating system and banking apps. There are likely to be many more updates in the near future as new vulnerabilities are uncovered and patched. And, as I’m sure you have been, you need to be ever vigilant for phishing attacks by email and SMS trying to obtain your banking credentials. The evolving threat landscape In the longer term, Mythos exposes the challenge that defence is much harder than attack. Software is one of the most complex products humanity builds. It is therefore almost impossible to ensure it is bug-free. That puts us in an unending race against the “bad guys” to uncover and fix faults before they get exploited. For example, with significant fanfare, the European Union just released its age verification app, designed to be a cornerstone to the emerging laws on access to social media, pornography and other age-restricted content. However, within hours, security experts found cyber vulnerabilities that underage users could easily exploit. In the most critical settings, we can try to prove mathematically that our software is bug-free. For instance, the Beneficial AI Foundation just announced an ambitious “moonshot” project to prove that the popular messaging app Signal is bug-free and protects privacy as claimed. But such efforts are the exception today rather than the norm. Perhaps further advances in AI could soon help reverse this. Toby Walsh receives funding from the Australian Research Council for a Laureate Fellowship on trustworthy AI.

The app lets users share disappearing photos with their friends that can be viewed only once and remain available for 24 hours.

<p><a href="https://openai.com/index/introducing-gpt-5-5/">GPT-5.5 is out</a>. It's available in OpenAI Codex and is rolling out to paid ChatGPT subscribers. I've had some preview access and found it to be a fast, effective and highly capable model. As is usually the case these days, it's hard to put into words what's good about it - I ask it to build things and it builds exactly what I ask for!</p> <p>There's one notable omission from today's release - the API:</p> <blockquote> <p>API deployments require different safeguards and we are working closely with partners and customers on the safety and security requirements for serving it at scale. We'll bring GPT‑5.5 and GPT‑5.5 Pro to the API very soon.</p> </blockquote> <p>When I run my <a href="https://simonwillison.net/tags/pelican-riding-a-bicycle/">pelican benchmark</a> I always prefer to use an API, to avoid hidden system prompts in ChatGPT or other agent harnesses from impacting the results.</p> <h4 id="the-openclaw-backdoor">The OpenClaw backdoor</h4> <p>One of the ongoing tension points in the AI world over the past few months has concerned how agent harnesses like OpenClaw and Pi interact with the APIs provided by the big providers.</p> <p>Both OpenAI and Anthropic offer popular monthly subscriptions which provide access to their models at a significant discount to their raw API.</p> <p>OpenClaw integrated directly with this mechanism, and was then <a href="https://www.theverge.com/ai-artificial-intelligence/907074/anthropic-openclaw-claude-subscription-ban">blocked from doing so</a> by Anthropic. This kicked off a whole thing. OpenAI - who recently hired OpenClaw creator Peter Steinberger - saw an opportunity for an easy karma win and announced that OpenClaw was welcome to continue integrating with OpenAI's subscriptions via the same mechanism used by their (open source) Codex CLI tool.</p> <p>Does this mean <em>anyone</em> can write code that integrates with OpenAI's Codex-specific APIs to hook into those existing subscriptions?</p> <p>The other day <a href="https://twitter.com/jeremyphoward/status/2046537816834965714">Jeremy Howard asked</a>:</p> <blockquote> <p>Anyone know whether OpenAI officially supports the use of the <code>/backend-api/codex/responses</code> endpoint that Pi and Opencode (IIUC) uses?</p> </blockquote> <p>It turned out that on March 30th OpenAI's Romain Huet <a href="https://twitter.com/romainhuet/status/2038699202834841962">had tweeted</a>:</p> <blockquote> <p>We want people to be able to use Codex, and their ChatGPT subscription, wherever they like! That means in the app, in the terminal, but also in JetBrains, Xcode, OpenCode, Pi, and now Claude Code.</p> <p>That’s why Codex CLI and Codex app server are open source too! 🙂</p> </blockquote> <p>And Peter Steinberger <a href="https://twitter.com/steipete/status/2046775849769148838">replied to Jeremy</a> that:</p> <blockquote> <p>OpenAI sub is officially supported.</p> </blockquote> <h4 id="llm-openai-via-codex">llm-openai-via-codex</h4> <p>So... I had Claude Code reverse-engineer the <a href="https://github.com/openai/codex">openai/codex</a> repo, figure out how authentication tokens were stored and build me <a href="https://github.com/simonw/llm-openai-via-codex">llm-openai-via-codex</a>, a new plugin for <a href="https://llm.datasette.io/">LLM</a> which picks up your existing Codex subscription and uses it to run prompts!</p> <p>(With hindsight I wish I'd used GPT-5.4 or the GPT-5.5 preview, it would have been funnier. I genuinely considered rewriting the project from scratch using Codex and GPT-5.5 for the sake of the joke, but decided not to spend any more time on this!)</p> <p>Here's how to use it:</p> <ol> <li>Install Codex CLI, buy an OpenAI plan, login to Codex</li> <li>Install LLM: <code>uv tool install llm</code> </li> <li>Install the new plugin: <code>llm install llm-openai-via-codex</code> </li> <li>Start prompting: <code>llm -m openai-codex/gpt-5.5 'Your prompt goes here'</code> </li> </ol> <p>All existing LLM features should also work - use <code>-a filepath.jpg/URL</code> to attach an image, <code>llm chat -m openai-codex/gpt-5.5</code> to start an ongoing chat, <code>llm logs</code> to view logged conversations and <code>llm --tool ...</code> to <a href="https://llm.datasette.io/en/stable/tools.html">try it out with tool support</a>.</p> <h4 id="and-some-pelicans">And some pelicans</h4> <p>Let's generate a pelican!</p> <div class="highlight highlight-source-shell"><pre>llm install llm-openai-via-codex llm -m openai-codex/gpt-5.5 <span class="pl-s"><span class="pl-pds">'</span>Generate an SVG of a pelican riding a bicycle<span class="pl-pds">'</span></span></pre></div> <p>Here's <a href="https://gist.github.com/simonw/edda1d98f7ba07fd95eeff473cb16634">what I got back</a>:</p> <p><img src="https://static.simonwillison.net/static/2026/gpt-5.5-pelican.png" alt="It is a bit mangled to be honest - good beak, pelican body shapes are slightly weird, legs do at least extend to the pedals, bicycle frame is not quite right." style="max-width: 100%;" /></p> <p>I've seen better <a href="https://simonwillison.net/2026/Mar/17/mini-and-nano/#pelicans">from GPT-5.4</a>, so I tagged on <code>-o reasoning_effort xhigh</code> and <a href="https://gist.github.com/simonw/a6168e4165a258e4d664aeae8e602cc5">tried again</a>:</p> <p>That one took almost four minutes to generate, but I think it's a much better effort.</p> <p><img src="https://static.simonwillison.net/static/2026/gpt-5.5-pelican-xhigh.png" alt="Pelican has gradients now, body is much better put together, bicycle is nearly the right shape albeit with one extra bar between pedals and front wheel, clearly a better image overall." style="max-width: 100%;" /></p> <p>If you compare the SVG code (<a href="https://gist.github.com/simonw/edda1d98f7ba07fd95eeff473cb16634#response">default</a>, <a href="https://gist.github.com/simonw/a6168e4165a258e4d664aeae8e602cc5#response">xhigh</a>) the <code>xhigh</code> one took a very different approach, which is much more CSS-heavy - as demonstrated by those gradients. <code>xhigh</code> used 9,322 reasoning tokens where the default used just 39.</p> <h4 id="a-few-more-notes-on-gpt-5-5">A few more notes on GPT-5.5</h4> <p>One of the most notable things about GPT-5.5 is the pricing. Once it goes live in the API it's <a href="https://openai.com/index/introducing-gpt-5-5/#availability-and-pricing">going to be priced</a> at <em>twice</em> the cost of GPT-5.4 - $5 per 1M input tokens and $30 per 1M output tokens, where 5.4 is $2.5 and $15.</p> <p>GPT-5.5 Pro will be even more: $30 per 1M input tokens and $180 per 1M output tokens.</p> <p>GPT-5.4 will remain available. At half the price of 5.5 this feels like 5.4 is to 5.5 as Claude Sonnet is to Claude Opus.</p> <p>Ethan Mollick has a <a href="https://www.oneusefulthing.org/p/sign-of-the-future-gpt-55">detailed review of GPT-5.5</a> where he put it (and GPT-5.5 Pro) through an array of interesting challenges. His verdict: the jagged frontier continues to hold, with GPT-5.5 excellent at some things and challenged by others in a way that remains difficult to predict.</p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/openai">openai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/chatgpt">chatgpt</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/llm-pricing">llm-pricing</a>, <a href="https://simonwillison.net/tags/pelican-riding-a-bicycle">pelican-riding-a-bicycle</a>, <a href="https://simonwillison.net/tags/llm-reasoning">llm-reasoning</a>, <a href="https://simonwillison.net/tags/llm-release">llm-release</a>, <a href="https://simonwillison.net/tags/codex-cli">codex-cli</a>, <a href="https://simonwillison.net/tags/gpt">gpt</a></p>

Noscroll wants to cure doomscrolling with an AI bot that reads the internet for you.

Kyle Jahner / Bloomberg Law: Music publishers including UMG, Warner Music, and Sony drop a copyright suit against Verizon following a SCOTUS decision limiting ISP liability in Cox's suit  —  Publishers representing the bulk of the music industry dropped a copyright lawsuit against Verizon Communications Inc. over music piracy …

Sean Plankey has requested to withdraw his name to run the U.S. cybersecurity agency after a tumultuous year of chaotic temporary leadership.

In this week’s episode of Uncanny Valley, we talk about Tim Cook’s legacy as CEO at Apple and what his long-rumored departure means for the future of one of the world's biggest companies.

Tom Warren / The Verge: Xbox CEO Asha Sharma and Chief Content Officer Matt Booty detail their “return of Xbox” strategy, including daily active players as its “new north star”  —  Microsoft's new Xbox leader starts to talk strategy for the company's gaming business.

Sydney Bradley / Business Insider: Instagram launched Instants, an app for sharing disappearing photos, in Italy this week, after rolling out an Instants feature in its main app in some regions  —  - Instagram launched a new app in Italy this week called “Instants.”  — A hodgepodge of apps like Snapchat, Locket, and BeReal, the app is meant to be “low-pressure.”

OpenAI: OpenAI says “GPT-5.5 matches GPT-5.4 per-token latency in real-world serving, while performing at a much higher level of intelligence”  —  A new class of intelligence for real work  —  We're releasing GPT-5.5, our smartest and most intuitive to use model yet …

Anthropic: Anthropic says it has fixed three causes of recent Claude Code quality issues: reduced default reasoning, a caching bug, and a system prompt to reduce verbosity  —  We traced recent reports of Claude Code quality issues to three separate changes.  Here's what happened and what we're changing.

Artisan may be known for their bold “Stop Hiring Humans” campaign but the reality is every founder needs to assemble the right team if they want to scale.

Discussion | Link

Discussion | Link

OpenAI says its latest model offers increased capabilities across a broad variety of categories.

Microsoft is offering first-ever retirement buyouts to some US employees as AI-driven data center spending grows and tech layoffs continue. The post Microsoft Offers First-Ever Retirement Buyouts to Thousands of Employees appeared first on TechRepublic.

CS 153 has gone viral on the Palo Alto campus—and on X. Not everyone is happy about it.

According to an internal memo, Meta plans to begin its sweeping layoffs on May 20.