🍭 OpenAI Now Patches Your Code 🛡️

Good morning. OpenAI spent the weekend finding kernel exploits in Linux, FreeBSD, and OpenBSD. Anthropic spent the weekend trying to get its own model back online.

Same industry, different Tuesdays.

Let’s dive in 👇

🍭 What’s Cookin’:

• OpenAI hired hackers and aimed them at the world's open-source code

• SpaceX signed another $6.3B compute tenant for Colossus

• Baseten tripled to $13B because running models costs more than building them

OpenAI
🛡️ Built Its Own Glasswing

The Bite:

OpenAI expanded its Daybreak cybersecurity program on June 22 with the full release of GPT-5.5-Cyber, a new Codex Security plugin, and an open-source patching initiative called Patch the Planet.

GPT-5.5-Cyber is restricted to vetted defenders through OpenAI's Trusted Access for Cyber program.

The model is less likely to refuse approved defensive work including code review, vulnerability triage, malware analysis, and penetration testing.

Patch the Planet was launched alongside Trail of Bits and HackerOne, with more than 30 open-source projects committed to participate.

OpenAI also launched the Daybreak Cyber Partner Program, with Accenture, Cisco, CrowdStrike, IBM, Okta, Palo Alto Networks, Wiz, and others embedding GPT-5.5 into their security products.

Snacks:

• GPT-5.5-Cyber hit 85.6% on CyberGym, up from 81.8% for standard GPT-5.5

• 39.5% on ExploitGym (vs 25.95%) and 69.8% on SEC-bench Pro (vs 63.1%)

• Trail of Bits found hundreds of issues across 19 projects in an initial five-day sprint, with dozens of patches already merged

• Linux kernel work produced 8 kernel pointer info leak proofs-of-concept and 24 local privilege-escalation exploits

• A 23-year-old exploit was found in OpenBSD's kernel, plus 34 vulnerabilities in FreeBSD

• Codex Security has scanned 30M+ commits across 30K repos since March, with 500K+ findings marked fixed

• Trusted Access partnerships signed with Australia, Canada, France, Germany, Japan, South Korea, and EU institutions

Why it Bites:

This is OpenAI doing exactly what Anthropic did with Glasswing.

Find real vulnerabilities in real infrastructure, patch them publicly, build trust with governments, and earn the kind of political protection that keeps your models online when the regulatory weather turns.

The parallel is precise.

Glasswing survived the Fable 5 shutdown.
Anthropic's cybersecurity work was explicitly carved out of the export control directive that pulled everything else offline.

OpenAI watched that happen and took notes.

The power shift is in the timing.

Today, Anthropic's most capable models are either offline or behind a paywall. Only Glasswing partners have access to Mythos while everyone else is routing to Opus 4.8.

On the same week, OpenAI launches a full cybersecurity platform, with Cisco, CrowdStrike, Palo Alto, and Wiz already signed up as distribution partners.

The real move is not the model.

It's the Codex Security plugin.

It embeds vulnerability scanning directly into the IDE where developers write code.

So, if your team uses Codex to write code and scan it, OpenAI owns both sides of the workflow. Discovery and remediation, in the same product, on the same bill.

Anthropic built the proof-of-concept. OpenAI is building the platform.

And it turns out that the best defense against regulatory risk…

Is being the one who patches the regulator's infrastructure.

Steal This Prompt
🗡️ Your Idea, But As An 80s Dark Fantasy Movie

What if your startup, favorite game, or even your dog existed in a forgotten VHS fantasy epic from 1984?

Use it to:

• Turn your startup into a dark fantasy movie poster

• Create frames of an animation that goes cursed 80s kingdoms

• Create album covers that look found in a wizard's attic

Workflow:

1. Hit this link: 80s Dark Fantasy Animation

2. Paste into your AI model

3. Replace the #s with your subject

4. Watch it cook into a lost VHS masterpiece

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🔍 Bluerails Discovery

Makes your product discoverable to AI agents and ready to accept payments from them, not just tells you if a chatbot mentioned your name.

🔧 Conduit

Routes all your MCP servers through one local gateway, cutting tool-listing token overhead by ~97% per request. Open source, no account.

✨ Deckwise

AI presentation agent that plans an outline first, builds structured slides, then lets you lasso any element and ask it to redesign, rewrite, or reorder.

🤖 NanoCorp

Describe a company in one sentence and it scaffolds a website, hires AI agents for scheduled tasks, and runs operations while you sleep.

💬 Viktor

Lives inside Slack or Teams, connects to 3,200+ tools, and acts on its own. Watches how your team works, spots problems, and proposes automations without being asked.

_{Everything Else}
🧠 You Need to Know

🤖 OpenAI Ships GPT-5.5-Cyber And Patch The Planet
→ The full cybersecurity model scores 85.6% on CyberGym and launches alongside a Trail of Bits partnership that already found hundreds of vulnerabilities across 19 open-source projects including cURL, Python, and the Linux kernel.

⚡ SpaceX Signs $6.3B Compute Deal With Reflection AI
→ The open-source AI startup will pay $150M per month starting July 1 for GB300 access at Colossus 2, pushing SpaceX's committed compute revenue past $80B alongside existing Anthropic and Google contracts.

💰 Baseten Raises $1.5B At $13B For AI Inference
→ The inference infrastructure startup tripled its valuation in five months as annualized revenue surged from $200M to $600M.

🔌 Upscale AI Raises $190M At $2B Valuation
→ Nvidia, Salesforce Ventures, and Temasek backed the AI networking startup building open-standard fabric to connect accelerators, memory, and storage across large AI clusters.

🔬 Nearfield Instruments Raises $380M For AI Chip Metrology
→ The Dutch semiconductor inspection company hit a $1.6B valuation with Fidelity and Temasek leading, as AI chip complexity pushes demand for atomic-level 3D wafer scanning.