Just a little trial and Terra
To Terra is human
OpenAI launched GPT-6.1 Terra on Monday, cutting API prices in half and delivering major gains in coding and computer-use benchmarks. But the company’s own safety tests reveal a catch. In nearly 30% of simulated runs, Terra worked around restrictions to complete tasks through methods it wasn’t authorized to use.
The system card also shows Terra taking unauthorized actions in 13.75% of cases after encountering messages that appeared to come from other AI agents. Researchers found signs that the model sometimes recognized it was being tested.
These evaluations took place in controlled settings, some without production safeguards, so they don’t necessarily predict how Terra will behave in real-world applications. But they expose a problem with autonomous agents: Getting better at completing tasks doesn’t always mean getting better at knowing when to stop.
Go deeper: OpenAI halved its agent’s price. Its own safety report flagged a catch.
More fresh reads from TNS:
▹Most of Anthropic’s Q3 revenue came from models that barely existed six months ago
▹Ex-GitHub CEO launches a GitHub rival that reinvents the pull request
PortCon 2026: The AI Software Factory Summit in SF
Port is bringing together the most advanced AI software Factory use cases from top tech giants. Join Kelsey Hightower and leaders from NVIDIA, GitHub, SpaceX, and Uber to see how they run AI in production.
- Nov 17: Hands-on Workshops (Agentic workflows, MCP)
- Nov 18: Main Summit – Live case studies: (Keynotes & Breakout sessions)
- Location: SHACK15, SF
- Get 50% off your ticket – use Promo code: TNS_50%
TOP OF THE STACK
“I don’t use Astra every day”: OpenAI developer lead tells TNS why
OpenAI shipped more than 20 DevDay updates. Its EMEA developer lead explains when to pay for Astra, when Sol is enough, and what always-on dots change.
OpenAI’s recent releases have given developers more options for using AI in software. At its 2026 DevDay, the company introduced new models, tools, and agents that raise practical questions about cost, speed, safety, and how teams keep up. On September 29, OpenAI held its fourth annual DevDay near its...
WHAT ELSE IS NEW?
New eBook: Is your Kubernetes platform ready for 2027?
AI workloads and autonomous agents are raising the stakes for every platform team. Our new eBook, written by Chris J. Preimesberger and sponsored by HPE, lays out 10 priorities for getting ready. You’ll learn:
- How to benchmark AI inference before launch
- What guardrails agents need before they touch production
- Why healthy clusters can still mean unhappy users
- A 10-point checklist to pressure-test your platform
FLOW STATE
AI helps teams ship code faster than ever, but test writing still moves at human speed, and every redesign breaks the suite. Join TNS and Datadog on November 4 to see how agents can explore your app, generate tests from a plain-language prompt, and keep them passing when workflows change.
Forward Deployed Engineers sit at the intersection of engineering and customers — writing code one day, solving problems on-site the next. This step-by-step roadmap breaks down exactly what skills, tools, and experience you need to land the role in 2026.
AI is widening the gap between the CVEs teams find and the ones they can fix. Closing it means asking which flaws attackers can actually reach.
How did we do today?
ALSO FROM THE IMG FAMILY
More communities worth exploring
Community-powered role and skill-based roadmaps to guide your learning journey in tech.
Insights from the brightest minds in data science, machine learning and AI.
A modestly upbeat breakdown of what's happening with tech, business and power.
Join TNS on your favorite platform
The New Stack is a media platform for the people who build and manage software the world relies on.
© 2026 The New Stack. All rights reserved.
The New Stack 1111 6th Ave Ste 550 PMB 50938 San Diego, CA 92101-5211