OpenAI has established the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), an independent panel of prominent mathematicians tasked with advising on the review and dissemination of the company's mathematical research results.
OpenAI introduced GPT-6 Sol and Luna, designed for high speed and cost efficiency, offering 50% lower API costs than GPT-5.6 models.
Apple researchers have introduced "probe guidance," a new method for flow matching models that utilizes frozen internal states to provide guidance signals without requiring additional forward passes during inference.
OpenAI and Grab have introduced a regional skilling program, "GO Forward with AI," aiming to equip 30,000 drivers, delivery partners, and merchants with practical AI skills over the next two years.
Google Labs is launching CC, an experimental AI agent that uses its own Google account to help up to six family members collaborate on calendars, tasks, and household logistics.
As AI demand surges, semiconductors are evolving into critical strategic resources, influencing global politics and economic security.
A recent analysis reveals that Chinese open-weight models have surpassed American counterparts in downloads, benchmark performance, and academic research citations, signaling a shift in the global AI landscape.
In September 2026, Persol Business Process Design began offering 3 services to support the preparation and operation of foundations that allow AI and AI agents to securely utilize business data. The company provides comprehensive support for challenges such as insufficient data organization, permission management, rising costs, and aging existing infrastructure.
Anthropic and OpenAI have released new efficiency-focused models that lower the cost of agentic workloads. This shift enables large-scale autonomous tasks, such as codebase-wide audits, by combining reduced token pricing with higher processing efficiency.
OpenAI has launched an improved prompt caching system for the GPT-6 family, offering higher default cache hit rates and new developer tools to optimize persistent agent workflows.
Rabbit's new "agentic operating system," OS3, can run locally across multiple desktop platforms without requiring the company's R1 hardware.
Codex CLI v0.156.0 introduces a fullscreen TUI option, default voice conversations, and an analytics dashboard for monitoring usage and token totals.
Microsoft has disrupted EvilTokens, a subscription-based phishing-as-a-service platform that used AI to compromise over 12,000 Microsoft accounts across 10,000 organizations.
British Columbia is suing OpenAI, alleging the company failed to report a high-risk user whose ChatGPT use was linked to a deadly mass shooting at a Canadian secondary school.
Ten Claude Opus 5.5 agents produced C-HD, a new algorithm that improves upon existing asymptotic bounds for shortest paths in directed graphs with non-negative real weights, verified via Lean.
An internal Pentagon review reveals that excessive dependence on Palantir's Maven Smart System, combined with outdated intelligence, contributed to a U.S. missile strike that killed at least 123 children at an Iranian school.
Unreal Labs has released Unreal Agent, an asynchronous harness that reduces model overhead during tool management, achieving up to 40% cost savings compared to Codex.
OpenAI expands its GPT-6 lineup with GPT-6 Sol and GPT-6 Luna, offering advanced intelligence with significantly lower API prices and improved performance in coding and professional workflows.
Coverage Cat introduces an insurance brokerage service that combines AI-guided intake with licensed brokers to compare home, umbrella, auto, and renters insurance.
Claude Code v2.1.280 introduces Claude Opus 5.5 as the default model for Pro and Team Standard plans and brings several improvements to VS Code and Slack integrations.