AI Tech Weekly

## OpenAI Unveils Sora 2: Next-Generation Video + Audio Generative Model and Companion Social App

On September 30, 2025, OpenAI publicly released **Sora 2**, its second-generation multimodal AI system capable of generating short videos with synchronized dialogue, sound effects, and more physically accurate motion than its predecessor. Sora 2 represents a significant advancement in text-to-video technology, offering tighter audio-visual integration and improved temporal consistency over multiple shots, addressing common “hallucination” issues in earlier video models ([openai.com](https://openai.com/fi-FI/index/sora-2/?utm_source=openai)) ([arstechnica.com](https://arstechnica.com/ai/2025/10/openais-sora-2-lets-users-insert-themselves-into-ai-videos-with-sound/?utm_source=openai)).

Simultaneously with the model release, OpenAI launched an **invite-only iOS app**—also called Sora—in the U.S. and Canada. The app features a TikTok-style feed where users can generate and share AI-created clips, and supports a “cameos” feature allowing users to insert themselves (with consent) into AI scenes by uploading a one-time video and audio recording ([techcrunch.com](https://techcrunch.com/2025/09/30/openai-is-launching-the-sora-app-its-own-tiktok-competitor-alongside-the-sora-2-model/?utm_source=openai)) ([cnbc.com](https://www.cnbc.com/2025/09/30/openai-sora-ai-video-generator-us-canada.html?utm_source=openai)).

The rapid adoption and viral growth of the Sora app have spurred debate over copyright and misuse. The Wall Street Journal reported that OpenAI will require studios and rights holders to **opt out** to prevent their copyrighted characters and styles from appearing in Sora outputs—shifting the burden to IP owners rather than implementing an opt-in system. OpenAI also promised to develop **more granular controls** for rights holders, akin to existing image-generation opt-out mechanisms, although public figures’ likenesses remain protected by a stricter “opt-in” verification process ([investing.com](https://www.investing.com/news/stock-market-news/openais-new-sora-video-generator-to-require-copyright-holders-to-opt-out-wsj-reports-4262080?utm_source=openai)) ([businessinsider.com](https://www.businessinsider.com/vinod-khosla-sora-openai-ai-slop-2025-10?utm_source=openai)).

## Anthropic Launches Claude Sonnet 4.5: Benchmark Leader in Coding and Autonomous Agents

On September 29, 2025, Anthropic announced **Claude Sonnet 4.5**, positioning it as the world’s “best coding model,” with substantial gains in coding performance, scientific reasoning, and sustained autonomous operation for up to 30 hours on complex, multistep tasks ([reuters.com](https://www.reuters.com/business/retail-consumer/anthropic-launches-claude-45-touts-better-abilities-targets-business-customers-2025-09-29/?utm_source=openai)) ([axios.com](https://www.axios.com/2025/09/29/anthropic-claude-sonnet-coding-agent?utm_source=openai)). Internal tests showed Sonnet 4.5 could write, debug, and deploy a web application over a continuous 30-hour session, compared to seven hours with its predecessor ([reuters.com](https://www.reuters.com/business/retail-consumer/anthropic-launches-claude-45-touts-better-abilities-targets-business-customers-2025-09-29/?utm_source=openai)) ([arstechnica.com](https://arstechnica.com/ai/2025/09/anthropic-says-its-new-ai-model-maintained-focus-for-30-hours-on-multistep-tasks/?utm_source=openai)).

Anthropic emphasized **enterprise readiness**, integrating Sonnet 4.5 into the Claude API, Microsoft 365 Copilot, and its own VS Code extension. The release included the **Claude Agent SDK**, enabling developers to build custom AI agents leveraging Sonnet 4.5’s extended context, checkpointing, and file-creation capabilities (spreadsheets, presentations, documents) directly within conversations ([techcrunch.com](https://techcrunch.com/2025/09/29/anthropic-launches-claude-sonnet-4-5-its-best-ai-model-for-coding/?utm_source=openai)) ([techradar.com](https://www.techradar.com/ai-platforms-assistants/claude/anthropics-claude-sonnet-4-5-is-available-now-the-best-ai-model-in-the-world-for-real-world-agents-coding-and-computer-use?utm_source=openai)).

The model incorporates strengthened **safety guardrails** to reduce risky outputs, aligning with Anthropic’s focus on dependable performance in regulated industries. Early adopters reported marked improvements in long-context code tasks and agentic workflows, underscoring the trend toward generative systems that can autonomously manage extended, complex processes ([axios.com](https://www.axios.com/2025/09/29/anthropic-claude-sonnet-coding-agent?utm_source=openai)) ([techradar.com](https://www.techradar.com/ai-platforms-assistants/claude/anthropics-claude-sonnet-4-5-is-available-now-the-best-ai-model-in-the-world-for-real-world-agents-coding-and-computer-use?utm_source=openai)).

## Meta to Acquire RISC-V Chip Startup Rivos to Power In-House AI Silicon

On September 30, 2025, Meta confirmed plans to acquire **Rivos**, a Santa Clara–based startup specializing in RISC-V-based AI accelerators, to bolster its in-house **Meta Training and Inference Accelerator (MTIA)** project. Rivos was reportedly nearing a $2 billion valuation, and Meta has been one of its biggest customers, seeking to reduce long-term reliance on external providers like Nvidia ([reuters.com](https://www.reuters.com/business/meta-buy-chip-startup-rivos-ai-effort-source-says-2025-09-30/?utm_source=openai)) ([investing.com](https://www.investing.com/news/stock-market-news/meta-to-buy-chip-startup-rivos-for-ai-effort-source-says-4264307?utm_source=openai)).

Rivos’s expertise in full-stack RISC-V GPU design and advanced memory technologies (e.g., HBM3e) is expected to accelerate Meta’s custom silicon roadmap, potentially allowing MTIA chips to handle both training and inference more cost-efficiently. Industry experts view the deal as part of a broader shift among major tech firms toward **vertical integration** of AI infrastructure, aiming for greater performance, power efficiency, and supply-chain resilience ([tomshardware.com](https://www.tomshardware.com/tech-industry/artificial-intelligence/meta-reportedly-buying-risc-v-ai-gpu-firm-rivos-acquisition-to-bolster-dev-team-and-possibly-replace-nvidia-internally?utm_source=openai)) ([theaicitizen.com](https://theaicitizen.com/p/top-ai-tech-news-through-october-5th?utm_source=openai)).

## EPA to Expedite Chemical Reviews for AI Data Center Projects

Effective **September 29, 2025**, the U.S. Environmental Protection Agency announced it will **prioritize** Toxic Substances Control Act (TSCA) new-chemical reviews for applications intended for AI data center construction or manufacturing of related components. This directive is part of Executive Order 14318, “Accelerating Federal Permitting of Data Center Infrastructure,” aiming to clear backlogs and **reduce regulatory delays** that could hamper the rollout of large-scale AI compute facilities ([epa.gov](https://www.epa.gov/newsreleases/epa-prioritizes-review-new-chemicals-used-data-center-projects-supporting-american?utm_source=openai)) ([en.reach24h.com](https://en.reach24h.com/news/industry-news/chemical/us-epa-prioritizes-new-chemicals-review-for-data-center-projects?utm_source=openai)).

Under the policy, new chemical manufacturers can request **priority review** by submitting a cover letter via the Central Data Exchange, identifying the intended data center project. While prioritization begins with submissions received on or after September 29, the EPA reaffirmed its commitment to thorough safety assessments, balancing **innovation acceleration** with environmental and human health protections ([epa.gov](https://www.epa.gov/newsreleases/epa-prioritizes-review-new-chemicals-used-data-center-projects-supporting-american?utm_source=openai)) ([en.reach24h.com](https://en.reach24h.com/news/industry-news/chemical/us-epa-prioritizes-new-chemicals-review-for-data-center-projects?utm_source=openai)).

## Bipartisan AI Risk Evaluation Act Introduced in Senate

On September 29, 2025, Senators Josh Hawley (R-Mo.) and Richard Blumenthal (D-Conn.) introduced the **Artificial Intelligence Risk Evaluation Act**, a bipartisan bill establishing an **Advanced AI Evaluation Program** within the Department of Energy. The legislation would require developers of **advanced AI systems** to submit detailed product information for risk assessment—covering loss-of-control scenarios, weaponization potential, and civil-liberties impacts—before public deployment ([axios.com](https://www.axios.com/2025/09/29/hawley-blumenthal-unveil-ai-evaluation-bill?utm_source=openai)) ([meritalk.com](https://www.meritalk.com/articles/senators-propose-doe-led-oversight-program-for-advanced-ai-risks/?utm_source=openai)).

The DOE program would systematically collect empirical data on AI risk profiles and report annually to Congress, creating a foundational **federal oversight mechanism** for emergent AI technologies. Sponsors argue this approach ensures **transparency, accountability**, and national-security safeguards without stifling innovation, reflecting growing bipartisan consensus on the need for thoughtful AI governance ([axios.com](https://www.axios.com/2025/09/29/hawley-blumenthal-unveil-ai-evaluation-bill?utm_source=openai)) ([meritalk.com](https://www.meritalk.com/articles/senators-propose-doe-led-oversight-program-for-advanced-ai-risks/?utm_source=openai)).

## AI Accelerates Antibiotic Discovery for Crohn’s Disease and IBD

On October 3, 2025, a collaboration between MIT’s CSAIL and McMaster University researchers demonstrated the power of AI in drug discovery by rapidly identifying **enterololin**, a narrow-spectrum antibiotic targeting bacteria implicated in Crohn’s disease and inflammatory bowel disease (IBD). Using a generative AI model to analyze over 10,000 microbial compounds, the team mapped the antibiotic’s mechanism and validated its efficacy in preclinical models within **six months** at a cost of approximately \$60,000—compared to traditional timelines of years and tens of millions of dollars ([news.mit.edu](https://news.mit.edu/2025/ai-maps-how-new-antibiotic-targets-gut-bacteria-1003?utm_source=openai)) ([medicinehatnews.com](https://medicinehatnews.com/news/national-news/2025/10/05/ai-helps-ont-researcher-discover-breakthrough-antibiotic-treatment-for-bowel-disease/?utm_source=openai)).

Enterololin selectively suppresses Enterobacteriaceae strains linked to gut inflammation while preserving beneficial microbiome species, offering a **precision-medicine** approach that could revolutionize IBD treatment. This breakthrough exemplifies how AI can **compress discovery pipelines**, enhance mechanistic insights, and unlock novel therapeutics for complex diseases ([news.mit.edu](https://news.mit.edu/2025/ai-maps-how-new-antibiotic-targets-gut-bacteria-1003?utm_source=openai)) ([medicinehatnews.com](https://medicinehatnews.com/news/national-news/2025/10/05/ai-helps-ont-researcher-discover-breakthrough-antibiotic-treatment-for-bowel-disease/?utm_source=openai)).

---

These developments—from multimodal generative models and advanced coding AI to custom silicon strategies, regulatory reforms, and transformative biomedical applications—underscore the extraordinary pace and breadth of AI advancements over the past week (September 29–October 5, 2025).