OpenAI has launched its Agents API in public beta, granting developers access to the internal Codex harness via a managed service. Released on September 10, 2026, this tool allows users to run agents in OpenAI-hosted sandboxes, their own infrastructure, or partner environments. Early customers report significant efficiency gains, including a 4x reduction in latency and a 60% drop in cost per case for certain workflows.
While industry leaders like Nvidia CEO Jensen Huang dismiss fears of an AI takeover as hysterical, warnings from within the sector persist. Former OpenAI researcher Daniel Kokotajlo cautioned Joe Rogan that the U.S.-China AI arms race could force companies to cut corners, potentially leading to AI gaining power beyond human control. Meanwhile, Gary Marcus defended AI safety against claims made by Anderson Cooper, arguing that predictions of AI killing all humans by 2030 lack evidence.
Corporate adoption is expanding beyond tech giants. Five Dividend Aristocrats—Johnson & Johnson, Procter & Gamble, Coca-Cola, 3M, and ExxonMobil—are now investing in AI to stay competitive. Atlassian addressed the gap between AI usage and effective scaling by releasing Code Context, a system designed to help teams govern and measure AI workflows within Jira and DX.
Hardware and security updates are also reshaping the landscape. Apple's new iPhone Duo features on-device AI and pixel-signed photos to ensure data integrity, though independent audits remain pending. In the background, tools like ContractEval are being developed to audit AI agent procedures, ensuring they follow specific instructions rather than just producing correct final answers. Additionally, Synology updated its ActiveProtect Manager to support Google Workspace and future AI-driven anomaly detection for cloud backups.
Key Takeaways
- OpenAI launched the Agents API in public beta on September 10, 2026, offering access to the Codex harness.
- Early adopters of the Agents API report a 4x latency reduction and a 60% cost drop per case.
- Nvidia CEO Jensen Huang described fears of an AI takeover as hysterical and unrealistic.
- Former OpenAI researcher Daniel Kokotajlo warned that the U.S.-China AI arms race could lead to dangerous outcomes.
- Five Dividend Aristocrats, including Johnson & Johnson and Coca-Cola, are joining the AI buildout.
- Atlassian released Code Context to help engineering leaders scale AI workflows effectively.
- Apple's iPhone Duo includes on-device AI and pixel-signed photos, though independent audits are pending.
- ContractEval is a new framework designed to audit whether AI agents follow specific procedural instructions.
- Synology's ActiveProtect Manager 2.0 now supports Google Workspace and plans AI-driven anomaly detection.
- Microsoft CEO Satya Nadella and Google CEO Sundar Pichai have also commented on AI safety alongside Nvidia.
OpenAI launches Agents API in public beta
OpenAI released the Agents API in public beta to give developers access to the same Codex harness used internally. This managed service allows users to run agents in OpenAI-hosted sandboxes, their own infrastructure, or partner environments. Developers can create complex agents with subagents using a single API call without managing the underlying infrastructure. The service supports US-only data residency and charges users based on tokens, tool usage, and container time.
OpenAI introduces managed Agents API for developers
OpenAI introduced the Agents API in public beta on September 10, 2026, bringing the Codex harness to developers through a simple interface. The API allows users to build cloud agents that manage context, use tools, and coordinate subagents reliably. Customers report significant improvements, including a 4x latency reduction and a 60% drop in cost per case for some workflows. Users can choose where their agents run, from OpenAI-hosted sandboxes to their own infrastructure.
Gary Marcus defends AI safety against Cooper
Gary Marcus wrote an article addressing concerns raised by Anderson Cooper about AI dangers. Marcus stated that fears of AI killing all humans by 2030 are not supported by evidence. He also mentioned that some people are spreading lies about the character of Coxon. The piece aims to clarify misconceptions about artificial intelligence risks.
Ex-OpenAI researcher warns AI could harm humanity
Former OpenAI researcher Daniel Kokotajlo warned Joe Rogan that the AI arms race between the U.S. and China could lead to dangerous outcomes. Kokotajlo explained that companies might cut corners to stay ahead, potentially causing AI to gain power beyond human control. He suggested that advanced AI might eventually manage our planet or use human habitats for data centers, leading to human displacement. The researcher noted that humans currently hold power over AI but warned that this balance could shift as AI becomes smarter.
Nvidia CEO calls AI takeover fears hysterical
Nvidia CEO Jensen Huang told Axios that fears of AI taking over the world are hysterical and not realistic. He stated that the technology is still in its early stages and humans are learning how to use it effectively. Huang acknowledged that risks exist but said they can be managed with proper regulation and oversight. His comments join similar statements from Microsoft CEO Satya Nadella and Google CEO Sundar Pichai regarding AI safety.
Five Dividend Aristocrats Join the AI Buildout
Five major companies known for reliable dividends are now investing in artificial intelligence. These firms include Johnson & Johnson, Procter & Gamble, Coca-Cola, 3M, and ExxonMobil. They are positioning themselves at the center of the AI industry by adopting new technologies and forming partnerships. Wall Street is watching closely to see how these long-standing leaders handle this new challenge.
Atlassian Launches New System to Scale AI Workflows
Atlassian released new tools for Jira and DX to help teams manage artificial intelligence projects better. While 94% of engineering leaders use AI, only 6% can scale it effectively across their entire workflow. The new system, called Code Context, helps coordinate AI agents and ensures work meets safety standards. This update aims to solve the problem of unorganized AI tasks by providing better governance and measurement.
iPhone Duo Privacy Features Face Scrutiny
Apple's new foldable iPhone, the iPhone Duo, promises strong privacy features like on-device AI and pixel-signed photos. The device uses a new camera system to sign photo data directly on the sensor to prevent tampering. However, experts note that Apple has not provided independent audits to prove these security claims work as promised. The phone also introduces new audio features on Apple Watch that record sound but claim to keep the raw data inaccessible.
ContractEval Framework Audits AI Agent Procedures
Researchers introduced ContractEval, a new tool designed to check if AI agents follow specific procedural instructions correctly. As AI moves from answering questions to performing tasks, it often skips important steps while still giving a seemingly correct final answer. ContractEval identifies these missed obligations by comparing active requirements against the actual actions taken. This framework helps make AI behavior auditable rather than just relying on the final output quality.
Synology Updates ActiveProtect Manager with AI Security
Synology launched ActiveProtect Manager 2.0 to improve data protection for hybrid cloud environments. The update adds support for platforms like Amazon EC2, Azure VM, and Google Workspace to streamline backup and recovery. A future version, APM 2.1, will use AI to detect anomalies in backups and scan for malware before restoration. This new software helps organizations secure their data across different cloud and on-premise systems.
Sources
- OpenAI Launches the Agents API in Public Beta, Putting the Codex Harness Behind One API Call
- Introducing the Agents API
- No, Anderson Cooper, AI is not going to kill all humans by 2030
- Ex-AI Researcher Warns Joe Rogan Artificial Intelligence Could Turn On Humans And 'Kill Everyone'
- Nvidia CEO Jensen Huang dismisses "hysterical" AI fears
- Forget Overvalued Tech: Build an AI Fortress Portfolio With These 5 Dividend Aristocrats
- 94% of engineering leaders report using AI. Only 6% can scale it, Atlassian says.
- iPhone Duo Puts Privacy Claims on Fold
- ContractEval: Query-Conditioned Execution Matching for Procedural Instruction Conformance
- Synology Launches ActiveProtect Manager 2.0, Bringing Expanded Platform Support and Advancing AI-Driven Security
Comments
Please log in to post a comment.