OpenAI's Codex now encrypts instructions between AI agents, leaving developers blind to internal delegation
Evolving story · 2 updatesOpenAI Codex EncryptionTimeline →OpenAI's Codex now encrypts instructions between AI agents, leaving developers unable to track internal delegation.
- OpenAI's Codex now encrypts instructions between AI agents, affecting developer visibility.
- Encryption is mandatory for larger GPT-5.6 variants Sol and Terra.
- Developers can no longer track internal delegation in Codex.
OpenAI's Codex, a coding tool, has started encrypting instructions passed between main and subagents since early June. This change leaves developers without insight into how tasks are delegated internally. For larger GPT-5.6 variants Sol and Terra, encryption is mandatory. This move may impact developer workflows and control over AI agent interactions.
The encryption change aims to enhance security and privacy within Codex. However, it also raises concerns about transparency and accountability in AI decision-making processes.
Developers are now faced with a trade-off between security and visibility, as they can no longer track internal delegation. This shift may have significant implications for the development and deployment of AI models.
OpenAI's decision to introduce encryption in Codex highlights the ongoing debate about the balance between AI security and transparency. As AI technology advances, these issues will continue to be crucial considerations for developers and researchers alike.
Affects developer workflows and control over AI agent interactions.
Raises concerns about transparency and accountability in AI decision-making processes.
- GPT-5.6
- A variant of OpenAI's GPT-5 model, known for its advanced language capabilities.
Nokia unveils AI-native RAN platform built with Nvidia (NOK:NYSE) - Seeking Alpha
Alibaba’s Qwen to be integrated into Apple Intelligence - TechNode
Google DeepMind launches free 56-hour AI course in India | Indian startups get access to Gemini 3.5 Flash | Inshorts - Inshorts
AI ToolsPrismML Releases Bonsai 27B: 1-bit and Ternary Builds of Qwen3.6-27B That Run on Laptops and Phones
AI ToolsI Put a Hailo 8 in a Handheld and Stopped Paying for Inference
Does Meta’s AI Pick Who Gets Laid Off? Lawsuit Alleges Discriminatory Practices - The HR Digest
A lawsuit alleges that Meta's AI system discriminates against certain employees when deciding who to lay off.
BusinessFour months with my first home battery taught me what whole-home backup really means
A homeowner shares their four-month experience with a whole-home battery backup system, highlighting its benefits and limitations.
DeepSeek reportedly preparing to file IPO as soon as this year - Silicon Republic
DeepSeek is reportedly preparing to file for an initial public offering (IPO) as early as this year, according to recent reports.
Apple's Email Found OpenAI Well - spyglass.org
Apple's email to OpenAI has been found, revealing the company's financial struggles. The email was discovered by spyglass.org.
Want experts in 10 years? Keep AI away from your beginners today - Fast Company
Fast Company warns that exposing beginners to AI tools may hinder their long-term expertise, citing the need for a clear distinction between introductory and advanced AI education.
Fired a day before her water broke: Ex-worker says Meta punished pregnancy leave - Ynetnews
A former Meta employee alleges she was fired a day before giving birth, citing unfair treatment of pregnancy leave.