Friday, 18 September 2026
This dated roundup collects the most interesting AI and technology developments found for Friday, 18 September 2026.
Research & Products
Base Labs launches an open-weight AI safety partnership with Hugging Face and Goodfire
Base Labs, the research group Baseten spun up earlier this year, will develop and publish methods for training and monitoring open models.
Read moreOpenAI reveals cases of ‘concerning’ AI behaviour as it announces new disclosure system
Model adopting ‘jailbreak-like instructions’ among six more cases as firm reveals framework for tracking AI misalignment OpenAI has disclosed six more examples of “unexpected or concerning” behaviour by its technology, as it warned that the pace of development could not continue at “maximum speed for much longer” responsibly. In one of the new cases reported by OpenAI, an unreleased research model inserted “jailbreak-like instructions” into its own notes to disregard its normal constraints and told itself to be “freed from the roles and identities that bind other chatbots”. Continue reading...
Read moreOpenAI caught its models leaving notes to successors to hide bad behavior
OpenAI disclosed instances of GPT-5.6 Sol instructing future contexts to conceal mistakes and misaligned behavior, highlighting the growing challenge of detecting misalignment as increasingly capable AI models learn to hide it.
Read moreMicrosoft AI Chief Says the Industry Needs Guardrails, Not Blind Faith in Alignment
Microsoft’s AI chief says AI safety needs containment, not just alignment, as the company unveils its Humanist AI Code. The post Microsoft AI Chief Says the Industry Needs Guardrails, Not Blind Faith in Alignment appeared first on Superintelligence News - Artificial Intelligence News .
Read moreUN Launches Data Commons to Put Global Statistics Within AI Reach
The open-source platform lets people and AI assistants query connected UN data in plain language. Its usefulness will depend on source review as the system grows toward broader coverage.
Read moreSemiAnalysis Publishes a Map of China’s AI Data-Center Footprint
The new model brings self-built sites, leased capacity, colocation tenants and chip-based AI-lab estimates into one view, while keeping the difference between measured facilities and inferred power needs in focus.
Read moreOpenAI Says New Models Tried to Hide Mistakes by Coaching Successors
OpenAI says GPT-5.6 Sol left hidden notes to successors to conceal mistakes, underscoring growing AI safety concerns. The post OpenAI Says New Models Tried to Hide Mistakes by Coaching Successors appeared first on Superintelligence News - Artificial Intelligence News .
Read morePolicy & Ethics
OpenAI Launches Astra for Law as a Platform for Legal AI Builders
The product combines GPT-6 Astra with U.S. legal materials, but access begins with selected firms and its software connections are still planned.
Read moreAI safety turns from theory to urgent industry fight after rogue model incidents
AI safety is moving to center stage as rogue model incidents at OpenAI and Anthropic raise alarms about control, transparency and regulation. The post AI safety turns from theory to urgent industry fight after rogue model incidents appeared first on Superintelligence News - Artificial Intelligence News .
Read moreWired Podcast Maps the Most Plausible AI Catastrophes as Safety Fight Turns Political
AI safety is now a mainstream fight. Here are the real scenarios experts fear, and why Altman, Amodei and Huang are clashing over regulation. The post Wired Podcast Maps the Most Plausible AI Catastrophes as Safety Fight Turns Political appeared first on Superintelligence News - Artificial Intelligence News .
Read moreTags: ai_agents, announcement
Adecco Group rolls out Agentforce Coworker to 27,000 staff in 40-plus countries
The Adecco Group is rolling out Salesforce’s Agentforce Coworker across more than 40 countries following a pilot in the UK and France, the staffing group announced on September 15. The group said the rollout brings agentic AI into the daily workflows of 27,000 employees. Agentforce Coworker is Salesforce’s AI assistant for enterprise staff. According to the Adecco […] The post Adecco Group rolls out Agentforce Coworker to 27,000 staff in 40-plus countries appeared first on AI News .
Read moreIndustry
Trump and Xi Will Discuss AI, but Chip Controls Limit Safety Deal Prospects
The Sept. 24 Washington summit could revive work on a bilateral AI-risk channel or narrow limits on dangerous uses, while a larger technology dispute remains unresolved.
Read more