Good morning. It’s Wednesday, September seventeenth.
On today in tech history: In 1988the Morris Worm became the primary major web worm, exploiting buffer overflows and weak password habits. It led on to the creation of CERT/CC and seeded a long time of research into intrusion detection and network security.
You read. We listen. Tell us what you think that by replying to this email.
How 433 Investors Unlocked 400X Return Potential
Institutional investors back startups to unlock outsized returns. Regular investors must wait. But not anymore. Due to regulatory updates, some firms are doing things in a different way.
Take Revolut. In 2016, 433 regular people invested a mean of $2,730. Today? They got a 400X buyout offer from the corporate, as Revolut’s valuation increased 89,900% in the identical timeframe.
Founded by a former Zillow exec, Pacaso’s co-ownership tech reshapes the $1.3T vacation home market. They’ve earned $110M+ in gross profit up to now, including 41% YoY growth in 2024 alone. They even reserved the Nasdaq ticker PCSO.
Paid commercial for Pacaso’s Regulation A offering. Read the offering circular at invest.pacaso.com. Reserving a ticker symbol isn’t a guarantee that the corporate will go public. Listing on the NASDAQ is subject to approvals.
Today’s trending AI news stories
OpenAI debuts GPT-5 Codex, adds latest personalization hub and teenage safeguards
OpenAI has launched GPT-5 Codex, a specialized model tuned for “agentic coding” that may run software development tasks constantly for over seven hours, debugging, refactoring, and even launching browsers for live validation. It integrates directly into developer pipelines, from VS Code to GitHub, and outperforms GPT-5 on SWE-bench Verified with a 74.5% rating, while showing dramatic improvements in code refactoring.
Meanwhile, a brand new personalization hub pulls custom instructions, communication preferences, and memory into one page. It doesn’t add fresh capabilities, but it surely kills the messy scattering of settings, finally making ChatGPT less annoying to configure.
A screenshot of the upcoming changes. | Image: Sam Altman
Under mounting scrutiny, OpenAI is tightening the screws on safety. It is going to now restrict users under 18 from accessing sexual or self-harm content, even in fictional contexts, and roll out parental controls including linked accounts, blackout hours, and distress alerts. An age-estimation system will default to treating ambiguous cases as minors, with ID checks required in some regions.
The corporate can be quietly reassembling a robotics team after shelving the project in 2020. Job listings cite teleoperation, simulation, and hardware prototyping. A key hire, Stanford’s Chengshu Li, has deep experience in humanoid robot benchmarks. On the business side, OpenAI brought on ex-xAI CFO Mike Liberatore to administer large-scale financing and cloud infrastructure partnerships, a critical step because it scales operations under a $300 billion Oracle cloud deal. Read more.
Google’s AP2 builds trust rails for AI economies, from stablecoins to bank cards
Google is formalizing the leap from AI assistants to autonomous dealmakers with its latest Agent Payments Protocol (AP2), an open standard that lets agents not only seek for bargains but actually execute transactions. Backed by greater than 60 partners including Mastercard, PayPal, Coinbase, and Salesforce, AP2 lays out a typical framework across bank cards, bank transfers, and stablecoins.
Every purchase runs on cryptographically signed mandates: an “intent” mandate defines scope and budget, while a “cart” mandate finalizes the buy. Users can pre-authorize stricter rules for fully delegated purchases, while still leaving an auditable trail. To increase reach into crypto, Google has built x402 with Coinbase, MetaMask, and the Ethereum Foundation, enabling self-custody stablecoin payments.
The push lands as DeepMind researchers warn of “agent economies” AI-run markets where agents trade and coordinate at machine speeds beyond human oversight. They caution that without guardrails, these economies could create flash-crash–like spirals or entrench oligopolies of powerful AI systems. Their proposals include sandboxed currencies, fairness-first auctions, “mission economies” aligned to human values, and underscore the stakes. AP2 positions itself as one such guardrail: interoperable, transparent, and auditable by design.
Alphabet is putting £5 billion into UK data infrastructure to fuel DeepMind and renewable-powered, heat-recycling data centers, only to be one-upped the identical day by Microsoft, which pledged £22 billion ($30B) for a national AI supercomputer and broader infrastructure push. Read more.
Meta Connect 2025 kicks off with AI glasses, neural wristbands, but no latest Quest
Meta Connect 2025 lands this week. The headline act is “Hypernova,” Meta’s first Ray-Ban glasses with a heads-up display in the proper lens, dual cameras, and a built-in AI assistant. It’s paired with a neural-input wristband powered by surface electromyography (sEMG).
Meta and EssilorLuxottica’s upcoming full lineup, from the video. The HUD glasses with sEMG wristband are on the far left | Image via: UploadVR
The band reads micro-finger movements, letting users type on a desk, flick through menus, or call up Meta AI without touching a screen. Priced around $800 and weighing about 70 grams, Hypernova is less full-blown AR than a practical HUD, positioned as a trendy, always-on interface somewhat than a bulky headset.
Alongside Hypernova, Meta will introduce Oakley “Sphaera” glasses built for athletes with a wraparound lens and central camera, plus a third-generation Ray-Ban Meta line. A leaked video confirmed the branding “Ray-Ban Display,” signaling deep buy-in from EssilorLuxottica. Hardware aside, the event will feature updates on Horizon OS licensing to Lenovo and Asus, but no latest Quest VR headsets until 2026, when a light-weight puck-driven model arrives. Read more.
Watch the complete event here.
-
ARC-AGI cracked with Grok-4 driven program synthesis at 10x lower cost
-
Alibaba open-sources Tongyi DeepResearch, a 30B web agent rivaling OpenAI
-
YouTube turns up the warmth with GenAI to outpace TikTok and Spotify
-
Productions brings human-in-the-loop editing to ElevenLabs’ AI audio stack
-
DeepSeek evaluates AI models for ‘frontier risks’, source says, as China promotes safety
-
BEHAVIOR-1K is about to turn out to be for robotics what ImageNet was for computer vision
-
Grok 5 starts training in a number of weeks
-
Anthropic launches Claude AI integration in Xcode 26 for advanced coding assistance and app development
-
Nvidia to deploy 120,000 AI chips within the U.K. by end of 2026
-
Epoch report for DeepMind predicts AI in 2030
-
Orchids, the world’s first AI full stack engineer
-
Aechelon launches Project Orbion, first AI-enabled Earth digital twin
-
WorldLabs improves AI geometry generation for full-scale 3D environment creation
-
Watch: Robotics milestone as AGIBOT X2 lands Webster flips on command
-
Former Google DeepMind researchers secure $5 million seed round for brand spanking new company to bring algorithm-designing AI to the masses
-
Watch: China’s humanoid robot survives brutal “violence test”
-
Oracle expands Fusion HCM with 13 latest AI agents for HR automation
-
Silicon Valley bets big on ‘environments’ to coach AI agents
-
Notion is planning to introduce Personalized AI Agents on September 18 during their upcoming keynote
-
Y Combinator-backed Rulebase desires to be the AI co-worker for fintech
-
Microsoft favors Anthropic over OpenAI for Visual Studio Code
-
The Anti-ChatGPT: Thomson Reuters’ multi-agent system slashes 20-hour tasks to 10 minutes
-
Top bank reveals sheer scale of the machines coming for our jobs
-
JP Morgan: AI will spark ‘violent task churn’ within the economy, but even optimists could also be underestimating the productivity boom
-
Watch: Boston Dynamic’s Atlas now squat, lowers its back, legs to perform middle distance manipulation
-
Nothing plans to launch ‘first AI-native devices next 12 months’
-
Alibaba launches the world’s first AI-native map application with Qwen
-
Zoom’s CEO agrees with Bill Gates, Jensen Huang, and Jamie Dimon: A 3-day workweek is coming soon due to AI
-
These are the tasks Indeed’s latest CEO says HR leaders should hand over to AI agents
-
Online marketplace Fiverr to put off 30% of workforce in AI push
-
AI found to successfully predict proposal outcomes using ISS guidelines
5 latest AI-powered tools from around the net
arXiv is a free online library where researchers share pre-publication papers.
Your feedback is beneficial. Reply to this email and tell us how you think that we could add more value to this text.
Considering reaching smart readers such as you? To turn out to be an AI Breakfast sponsor, reply to this email or DM us on 𝕏!
