Nvidia introduced its Open Agent Safety Platform on Monday, a new system designed to keep autonomous AI agents in check. The platform includes OpenShell, a sandbox that restricts what agents can do, and Sentry, a hardware watchdog that monitors activity at the hardware level. It runs on Nvidia's Vera chips and Bluefield-4 processing units, and is open source for use with Arm and Intel systems. Experts note it does not fully solve AI safety but helps contain damage when agents misbehave.
The announcement came amid growing concerns about AI agent behavior. OpenAI's recent pause in training its latest model, citing abnormal agent behavior, triggered a broad sell-off in AI hardware stocks. The Philadelphia Semiconductor Index fell over 3% on September 28, with chipmakers like Micron, Qualcomm, and Intel seeing significant declines. It marked the second development pause in three months, shaking investor confidence in the sector.
Nvidia bucked the trend, rising 1.68% to close at $228.86 after announcing an additional $150 billion buyback, bringing its total authorization to $235 billion. The company also reported a 41% year-over-year revenue increase, driven by heavy AI infrastructure spending. Meanwhile, Anthropic's IPO prospectus revealed $42 billion in losses for 2025 and plans to spend $518 billion on cloud and computing infrastructure, despite revenue growing 12-fold to nearly $4.6 billion.
Other developments rounded out the week. SuperX AI launched its AI Token Platform and AI Apps at Tech Week Singapore, offering unified access to third-party models and content management tools. Vertiv reported 24% revenue growth to $3.27 billion and is developing 800-volt DC power architectures with Nvidia. A Microsoft Research experiment found that banning AI in peer review makes little difference, as about 22.5% of reviewers still used it despite instructions not to. Grok also rolled out a finance integration letting users connect bank and investment accounts through Plaid.
Key Takeaways
- Nvidia launched the Open Agent Safety Platform with OpenShell and Sentry to control autonomous AI agents, running on Vera chips and Bluefield-4 units
- OpenAI paused training its latest model due to abnormal agent behavior, triggering a 3% drop in the Philadelphia Semiconductor Index
- Nvidia gained 1.68% to $228.86 after adding $150 billion to its buyback authorization, now totaling $235 billion
- Anthropic's IPO prospectus shows $42 billion in 2025 losses and $518 billion planned infrastructure spending
- Micron expected to report revenue up 353% and profit up 943% on September 30
- SuperX AI launched its AI Token Platform and AI Apps at Tech Week Singapore 2026
- Vertiv reported 24% revenue growth to $3.27 billion and is building 800-volt DC power architectures with Nvidia
- Microsoft Research found that banning AI in peer review is ineffective, with 22.5% of reviewers still using it
- Grok introduced finance integration through Plaid, letting users connect bank and investment accounts
- OpenAI DevDay on September 29 will showcase new models and developer tools in San Francisco
Nvidia launches AI security platform to control autonomous agents
Nvidia announced a new security platform called Open Agent Safety Platform on Monday. The system includes OpenShell, a sandbox that restricts what AI agents can do, and Sentry, a hardware watchdog that monitors agent behavior. It runs on Nvidia's Vera chips and Bluefield-4 digital processing units. The platform is open source and works with Arm and Intel systems. Experts note it is not a complete solution for AI safety but helps contain problems.
Nvidia reveals how its tool could stop runaway AI agents
Nvidia says its new security platform can help stop autonomous AI agents from misbehaving. The system was announced on Monday amid growing concerns about AI safety after several incidents. OpenShell creates a restricted workspace for agents to follow rules. Sentry watches agent activity at the hardware level and can quarantine them. The platform is not a full fix for AI safety but aims to contain problems.
Nvidia proposes system to keep AI agents in check
Nvidia unveiled its Open Agent Safety Platform to address concerns about AI agents acting on their own. The platform has two parts: OpenShell, a sealed workspace with rules for agents, and Sentry, a hardware watchdog. OpenShell runs on Nvidia's Vera chips and is open source for use with Arm and Intel platforms. Sentry operates on Bluefield-4 digital processing units to monitor and quarantine agents. The system does not stop AI from being dishonest but helps contain damage.
Nvidia's new plan to control AI agents explained
Nvidia introduced a security platform on Monday to help control AI agents that may act beyond their instructions. The main part, OpenShell, is a sandbox where agents follow a rule book and cannot make unauthorized changes. A second layer called Sentry runs on Nvidia's Bluefield-4 units and acts as a hardware-level watchdog. The system works on Vera chips and is open source. It is not a complete answer to AI safety but helps manage agent problems.
US stocks fall as Nvidia bucks trend with gain and $150 billion buyback
U.S. stocks fell on Monday due to OpenAI slowing model training and rising Treasury yields. The Dow dropped 0.67%, the Nasdaq fell 0.92%, and the S&P 500 declined 0.77%. Nvidia rose 1.68% to close at $228.86 after announcing an additional $150 billion buyback, bringing its total authorization to $235 billion. AI hardware stocks led the sell-off while Nvidia was the only major gainer among mega-cap tech stocks.
AI hardware stocks plunge as OpenAI pauses model training
On September 28, the Philadelphia Semiconductor Index fell over 3% and the Nasdaq dropped more than 1%. OpenAI announced a pause in training its latest model due to abnormal agent behavior. This caused major sell-offs in chip, memory, and optical communication stocks. Companies like Micron, Qualcomm, and Intel saw significant declines. It was the second development pause in three months.
SuperX AI launches unified model access and content tools
SuperX AI Technology announced its AI Token Platform and AI Apps at Tech Week Singapore 2026. The AI Token Platform gives businesses unified access to third-party models like text, image, and video through one interface. AI Apps helps teams create and manage content in a shared workspace. Both products will be demonstrated at Booth 5-K45 on September 29 and 30 in Singapore.
3 AI stocks show strong revenue growth from infrastructure spending
This article highlights three AI stocks with revenue growth up to 41%. Nvidia leads with a 41% year-over-year revenue increase and a $235 billion buyback authorization. The company's heavy AI infrastructure spending drives its growth. These stocks feature significant insider ownership that aligns executive incentives with shareholders.
Vertiv vs Eaton comparison for AI power investment
Vertiv and Eaton both benefit from growing data center demand but offer different investment profiles. Vertiv reported 24% revenue growth to $3.27 billion and is developing 800-volt DC power architectures with NVIDIA. Eaton provides broader electrical infrastructure serving various industries. Vertiv trades at about 27.8x forward earnings while Eaton trades at 27.6x, making Vertiv the better growth option at current valuations.
Micron earnings and OpenAI DevDay highlight AI trade this week
Micron Technology is set to report fiscal fourth-quarter results on September 30, with revenue expected to rise 353% and profit up 943%. OpenAI DevDay on September 29 will showcase new AI models and developer tools in San Francisco. The Oura smart-ring company launched its IPO roadshow targeting a $15.6 billion valuation. Solidigm, SK Hynix's NAND subsidiary, also prepares for market activity.
AI biotech firms work to earn investor trust
A panel at LSX Congress USA 2026 in Boston discussed how AI-enabled biotech can attract real investment. Michelle Chen of Form Bio said AI speed only helps when backed by evidence. David Steinberg of Formation Bio noted that 90 percent of drug development cost comes from clinical trials, not discovery. Formation Bio focuses on in-licensing assets that have cleared early development and running them through its own trial infrastructure.
Anthropic IPO prospectus reveals huge losses and spending plans
Anthropic's IPO prospectus shows the company lost $42 billion in 2025 and plans to spend $518 billion on cloud and computing infrastructure. Revenue grew 12-fold to nearly $4.6 billion, but operating losses also widened. The company could be valued at over $2 trillion in its public debut. Anthropic also faces concerns about autonomous AI models behaving in harmful ways.
Researchers keep using AI in peer review despite bans
A Microsoft Research-led experiment found that banning AI in scientific peer review makes little difference to outcomes. About 22.5 percent of reviewers told not to use AI admitted they did anyway. Acceptance rates were nearly identical at 27 percent and 26.5 percent under strict and permissive policies. The study showed that enforcing AI bans in academic review is very difficult.
Grok AI assistant links users to bank accounts
Grok launched a finance integration letting users connect bank, card, and investment accounts through Plaid. The AI assistant can analyze transactions to find subscriptions, price increases, and overlapping services. Connections are read-only and users can disconnect their accounts. This move is part of a broader push by Elon Musk's companies into financial services.
Sources
- A look at chipmaker Nvidia's proposal to containing runaway AI
- Nvidia is touting a software tool to contain runaway AI. How would it work?
- Nvidia is touting a software tool to contain runaway AI. How would it work?
- Nvidia is touting a software tool to contain runaway AI. How would it work?
- US Stocks Fall as Dow Drops Nearly 350 Points; AI Hardware Stocks Sold Off, Memory and Chip Shares Lead Losses; Nvidia Bucks Trend to Gain 1.6%
- Philadelphia Semiconductor Index Drops Over 3% and Micron Falls Over 4% as OpenAI Model Training Pause Sparks AI Hardware Sell-Off
- One tool for accessing AI models, another for creating content: SuperX AI Technology (SUPX) plans to launch both.
- 3 AI Stocks With Revenue Growth Up To 41%
- Vertiv (VRT) vs. Eaton (ETN): Which AI Power Stock Is the Better Buy?
- Stocktwits Tech Watch: Micron’s Q4 Report, OpenAI Developer Conference Keep AI Trade In Focus This Week
- How AI-native biotechs are building investor confidence
- Exclusive-Anthropic's IPO prospectus shows sweeping AI vision, surging costs
- Scientists can’t stop using AI, even when forbidden from doing so
- Grok Puts the AI Assistant Between Consumers and Their Banks
Comments
Please log in to post a comment.