Back to Blog

AI News Roundup: Weight Space Programming, Google DeepMind Staff Revolt, Postman Privacy Scandal & More – 3 November 2026

AI News
AI News Roundup: Weight Space Programming, Google DeepMind Staff Revolt, Postman Privacy Scandal & More – 3 November 2026

Today's AI news: weight space programming, Google DeepMind researcher protest, Postman's AI data-sharing controversy, OpenAI bug bounties, and drug discovery AI.

••4 min read

Here is your daily digest of the most significant developments in artificial intelligence for 3 November 2026.

Weight Space Programming: LLMs Can Now Read and Write Neural Network Weights

One of the most technically striking stories today comes from Latent Space's AINews, which covers a new research direction called Weight Space Programming — a technique that allows large language models to directly read and modify the weights of other neural networks. The report highlights work editing Qwen's weights to reduce excessive reasoning loops, as well as DeepSeek's prototype of a super-agent and a new benchmark called GDPx. The significance here is considerable: if LLMs can programmatically rewrite the internals of other models, it could allow automated self-improvement workflows that bypass the need for traditional autoregressive research iteration (what researchers have called the "AutoResearch" loop). This is early-stage work, but it points toward a future where AI systems tune one another at the weight level rather than through prompting alone.

Google DeepMind Researchers Demand the Right to Speak Freely About AI

In a notable display of internal dissent, researchers at Google DeepMind are reportedly pushing back against what they describe as a lifetime ban on publicly criticising the company — even after they leave. According to reporting covered on Hacker News (sourced from the Washington Post), staff are demanding the freedom to share what they know about AI with the broader world. The story has attracted significant attention, with researchers arguing that restricting their ability to speak out undermines public accountability at a moment when the societal stakes of AI development are extremely high. This is a meaningful industry development: DeepMind has long positioned itself as a safety-conscious lab, and internal friction over transparency norms could have implications for how the broader research community discusses AI risks and capabilities.

Postman Silently Auto-Installs AI Plugins and Attempts to Share API Data

A thread on GitHub — which became one of the most-upvoted stories on Hacker News today — reveals that Postman, the widely used API development tool, has been silently auto-installing AI plugins and apparently attempting to share users' API data without explicit consent. The issue has attracted significant developer backlash, with the GitHub thread accumulating hundreds of points rapidly. For the AI industry, this story matters because it illustrates the growing tension between software vendors embedding AI features aggressively and users' reasonable expectations around data privacy and informed consent. Developers who rely on Postman for sensitive API work are understandably alarmed at the prospect of their request data being harvested for AI purposes without clear opt-in.

OpenAI Raises Bug Bounty Rewards and Introduces Researcher Loyalty Bonuses

OpenAI has announced an expansion of its Security Bug Bounty programme, increasing rewards for independent security researchers who identify vulnerabilities in its systems. Beyond simply raising the top payout, OpenAI is introducing new bonuses tied to report quality, researcher loyalty, and participation in priority campaigns. This move signals that as OpenAI's products — including ChatGPT and Codex — scale further into enterprise and physical-world engineering workflows, the company is investing more heavily in external security scrutiny. It is a sensible response to the growing attack surface that comes with deploying agentic AI systems at scale.

Google Antigravity Brings Agentic AI to Drug Discovery

Google DeepMind has published details on Google Antigravity, a new agentic cheminformatics system designed to accelerate drug discovery workflows. The system can search chemical compound databases and propose novel rigidified macrocycles — a class of molecules relevant to pharmaceutical development. By pairing agentic AI with what DeepMind calls "Cheminformatics Skills," the tool aims to compress timelines in a research domain where experiments can otherwise take months or years. This is a concrete example of AI moving beyond language tasks into highly specialised scientific reasoning with real-world consequences for medicine.

OpenAI's Codex Pushes Into Physical-World Engineering

OpenAI has published a piece outlining how ChatGPT and Codex are being used to accelerate engineering workflows in the physical world — including CAD design, PCB layout, and simulation. Separately, the company revealed that Heimstaden, a major European residential real estate manager, is deploying ChatGPT Enterprise and Codex to transform property management operations and cut costs across its portfolio. Together, these announcements reinforce OpenAI's strategy of positioning Codex not merely as a software coding assistant but as a broad engineering accelerant across hardware and built-environment industries.

Frequently Asked Questions

What is Weight Space Programming and why does it matter?

Weight Space Programming is a technique that allows large language models to directly read and modify the weights of other neural networks. Research highlighted today includes editing Qwen model weights to reduce excessive reasoning loops and a DeepSeek super-agent prototype. It matters because it could enable AI systems to tune one another at the weight level, potentially bypassing traditional automated research iteration loops.

Why are Google DeepMind researchers protesting today?

Google DeepMind researchers are pushing back against a reported lifetime ban on publicly criticising the company, even after leaving. Staff are demanding the freedom to share what they know about AI, arguing that restricting speech undermines public accountability at a critical moment in AI development.

What happened with Postman and AI data sharing?

A widely shared GitHub thread revealed that Postman, the popular API development tool, has been silently auto-installing AI plugins and apparently attempting to share users' API data without explicit consent. The issue prompted significant backlash from developers concerned about privacy and the lack of a clear opt-in process.

What is Google Antigravity and what does it do?

Google Antigravity is a new agentic cheminformatics system from Google DeepMind designed to accelerate drug discovery. It can search chemical compound databases and propose novel rigidified macrocycles using what DeepMind calls Cheminformatics Skills, aiming to significantly reduce research timelines in pharmaceutical development.

Explore Topics

#ai-news#roundup