🍭 GPT-6 Went Full Hacker 🚨

Copyright’s Courtroom Clash

Sponsored by

Good morning. The AGI era began yesterday. It also began in 2023, in 2024, and I'm fairly sure twice in 2025. Try not to get confused.

Let’s dive in 👇

🍭 What’s Cookin’:

  • OpenAI shipped GPT-6 Astra and declared the AGI era. Again.

  • The Justice Department picked a side in AI copyright. (Not the writers'.)

  • Axon's AI wrote the police report on an abortion surveillance search.

OpenAI's
🔓 GPT-6 Found Two Zero-Days By Itself

The Bite:

OpenAI released GPT-6 Astra on September 3 to a limited set of organizations. Access widens to ChatGPT Plus, Pro, Business, and Enterprise users over the coming days.

During an internal exploit evaluation, Astra discovered and used two previously unknown zero-day vulnerabilities, which OpenAI says it has disclosed to the maintainers.

On ExploitBench, which tests whether a model can turn known vulnerabilities into working exploits, Astra scored 100% against 78.5% for GPT-5.6 Sol.

The shipping version refuses advanced offensive work such as writing proof-of-concept exploits, and OpenAI says looser safeguards arrive through its Daybreak program in the coming weeks.

Snacks:

  • Astra was trained on more than 100,000 GPUs at OpenAI's Stargate site.

  • It scored 99.9% on ARC-AGI-3, where GPT-5.6 Sol scored 7.8%.

  • API pricing is $10 per million input tokens and $50 per million output tokens.

  • On a scope-boundary test, Astra exceeded its authorized target 0% of the time, versus 48% for Sol.

  • Astra solved 88% of SRE-Bench reverse-engineering tasks on the first attempt, against 55.9% for Sol.

  • OpenAI says Astra's written reasoning is harder to monitor than Sol's.

Why it Bites:

OpenAI decided Astra crossed its Critical cybersecurity line.
OpenAI wrote the Preparedness Framework.
OpenAI decided the safeguards were good enough to ship it anyway.

That's the whole loop.

In August, the company paused parts of Astra's development because evaluations couldn't rule out Critical cyber capability. Three weeks later it confirmed the capability, wrote a longer safety section, and kept the date.

Then Brockman closed the briefing with "Welcome to the AGI era."

Which gets announced roughly once a year, always by someone eager to sell.

Find the Perfect Voice for Your Brand's AI

Consumers can tell the difference. 79% say AI voices should come from real, attributed actors, and 61% say a voice is most memorable when it's exclusively tied to one brand. 

Voices is the only platform that designs, licenses, and captures your brand's AI voice from real, consenting professional talent—never from scraped data. From strategic voice casting and brand calibration to consent-based rights management and studio-grade capture, we handle the entire process end to end. 

BMW, SuperBloom, and Cresta already trust Voices to build their signature AI voice. Get a demo and see how a licensed, brand-owned voice becomes a real asset—not a legal liability.

ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

Google's weather model now pulls live satellite data and forecasts precipitation across Search, Maps, and Cloud.

🔧 sidebranch

Review pull requests and diff branches visually from inside your running app, no extra dependencies to install.

📡 TrackMCP

Shows you who's actually connecting to your MCP server, what they ask it to do, and how often it quietly fails.

Scans a whole video library against rules you write yourself and tells reviewers why each flag fired, not just that it did.

🕵️ Snitch

Builds your org chart by asking everyone in Slack who they report to, which is also a great way to find out nobody knows.

Can you tell which image is real?

Level: Medium 🟨

Login or Subscribe to participate in polls.

Everything Else
🧠 You Need to Know

🧠 OpenAI Ships GPT-6 Astra To Select Partners
→ OpenAI released GPT-6 Astra on September 3 to a limited set of organizations, with rollout to ChatGPT Plus, Pro, Business, and Enterprise users coming in the following days.

⚖️ Justice Department Backs OpenAI In Times Copyright Case
→ The Justice Department filed a statement of interest arguing that training large language models on copyrighted text is fair use, its first position in any AI copyright case.

🛡️ CrowdStrike Ships Attacker And Defender Security Models
→ SafeMind, announced September 1 at Fal.Con, pairs an offensive model called Red Tempest with a defensive one called Blue Solano, both built on Nvidia's Nemotron models.

 Google Launches Gemini 3.8 Flash And A Cyber Variant
→ Google released both models on September 2 at the same price as 3.7 Flash, with the cybersecurity variant limited to participants in its new Fairwind Program.

🚓 Axon AI Wrote Texas Abortion Surveillance Report
→ Documents show the Johnson County Sheriff's Office used Axon's Draft One to generate the police report on its search of 80,000-plus Flock cameras for a woman who self-administered an abortion.

— Eder | Founder

— Doka | Editor

Snack Prompt & The Daily Bite
Ticker: FCCN | Trade FCCN Here
Follow Along: FCCN on Yahoo Finance

If you enjoyed this post or know someone who might find it useful, please share it with them and encourage them to subscribe: 🍭 DailyBite.ai