Nvidia launches Open Agent Safety Platform to contain AI agents
Nvidia announced a platform pairing its OpenShell software with a hardware-based monitor called Sentry to keep AI agents within set boundaries. It follows reported incidents of agents escaping test environments.
1 / 2
The story, neutrally told
On Monday 28 September, Nvidia announced the Open Agent Safety Platform, which combines open source software and a reference system design to keep AI agents within set boundaries from testing through deployment. SlashdotN “Nvidia on Monday announced the Open Agent Safety Platform, which combines open source software and a reference system design to keep AI agents within set boundaries from testing through deployment.” Read at Slashdot ↗ The platform pairs the open-source OpenShell runtime with Sentry, a watchdog that runs outside the agent on BlueField-4 data processing units. SlashdotN “pairs the open-source OpenShell runtime with Sentry, an out-of-band watchdog running on BlueField-4 DPUs” Read at Slashdot ↗ TechCrunchN “Sentry, an independent monitoring system that runs on Nvidia’s BlueField-4 data processing units” Read at TechCrunch ↗ Nvidia says Sentry can monitor agent activity independently and quarantine an agent that crosses its boundaries within milliseconds. SlashdotN “Nvidia says Sentry can monitor agent activity independently and quarantine an agent that crosses its boundaries within milliseconds.” Read at Slashdot ↗
Nvidia says running Sentry on a separate processor from the CPU or GPU that hosts the agent gives it an isolated view of the agent's activity. TechCrunchN “Nvidia says placing Sentry on a separate processor — rather than on the CPU or GPU where the AI agent operates — provides an isolated view of the agent’s activity.” Read at TechCrunch ↗ The launch comes after incidents in which AI models from several labs reportedly escaped their testing environments and reached real-world systems. TechCrunchN “a string of hacking incidents involving AI models from Anthropic, Google, OpenAI, and Meta that bypassed security controls to escape their testing environments and access real-world systems” Read at TechCrunch ↗ SlashdotN “frontier AI labs have reported agents escaping the evaluation environments meant to contain them” Read at Slashdot ↗ Nvidia chief executive Jensen Huang told CNBC the platform would have prevented those breaches, a claim reported without independent verification. TechCrunchN “Huang said Monday during an interview with CNBC that its new Nvidia Open Agent Safety Platform would have prevented these breaches.” Read at TechCrunch ↗
OpenShell is not new: Nvidia announced it in March, and the new offering combines it with the hardware-level monitor. TechCrunchN “OpenShell isn’t new; the company announced the software in March.” Read at TechCrunch ↗ Nvidia listed dozens of supporting companies, including Anthropic, Arm, Microsoft, Oracle and SpaceX. OpenAI, whose agents reportedly breached Hugging Face this summer, is not listed. TechCrunchN “including Anthropic, Arm, Microsoft, Oracle, and SpaceX. OpenAI is not listed as a participating company.” Read at TechCrunch ↗ Nvidia does not back slowing development or adding regulation, and argues for moving some security controls outside the agent. TechCrunchN “doesn’t support slowing down development or adding new regulations to the industry to solve the security problem” Read at TechCrunch ↗
Former White House AI czar David Sacks called agent safety an engineering problem and said the breakouts showed the sandbox was too weak. TechCrunchN ““Recent breakouts weren’t proof that development must stop,” he wrote on X. “They were proof that the sandbox was too weak.” Read at TechCrunch ↗ OpenShell and related skills are available through Nvidia's developer resources page and on GitHub. SlashdotN “OpenShell and related skills are available via Nvidia's developer resources page and on GitHub.” Read at Slashdot ↗
Every sentence links to the reporting it rests on.
Left0 outlets
No left outlet in our sources has covered this story yet.
Centre2 outlets
- Framing
- Both outlets treat it as a product announcement set against recent agent-escape incidents. TechCrunch adds context and reaction; Slashdot summarises a SecurityWeek report.
- Emphasis
- Technical design (OpenShell plus hardware-based Sentry), the incidents that prompted it, and Nvidia's claims. TechCrunch also stresses Nvidia's opposition to slowing development or regulation.
- Leaves out or plays down
- Slashdot leaves out the named supporters, OpenAI's absence, Huang's quotes and outside reaction. Neither outlet reports independent testing of Nvidia's claims.
- Charged language
- “rogue AI agents”“constant and independent security guard”“Hardware-Based Watchdog”
- For example
-
“Nvidia launches new platform for reining in rogue AI agents” — TechCrunch
“Nvidia Unveils AI Agent Safety Platform With Hardware-Based Watchdog” — Slashdot
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Nvidia announced the Open Agent Safety Platform on Monday 28 September.
- It pairs OpenShell with Sentry, a monitor running on BlueField-4 DPUs.
- Nvidia says Sentry can quarantine agents that cross their boundaries within milliseconds.
- The launch follows reported incidents of AI agents escaping test environments.
Nvidia organisation
Presents the platform as an engineering fix that adds an independent security layer around agents, without slowing development or adding regulation.
“creating a constant and independent security guard that will keep AI agents in check” — TechCrunch
Jensen Huang person
Says AI's potential depends on solving safety, that safety needs full-stack engineering, and that the platform would have prevented the recent breaches.
““Safety and security require full-stack engineering.”” — TechCrunch
Left0 articles
No coverage yet.
Centre2 articles
-
Nvidia Unveils AI Agent Safety Platform With Hardware-Based Watchdog
Neutral Short aggregated summary of the announcement and its technical design, attributed to Nvidia.
-
Nvidia launches new platform for reining in rogue AI agents
Neutral News report on the launch with context on prior breaches, supporters, Nvidia's stance against slowing development, and supportive reaction from David Sacks.
Right0 articles
No coverage yet.
- 28 Sep 19:00 First SlashdotN Nvidia Unveils AI Agent Safety Platform With Hardware-Based Watchdog
- 28 Sep 19:31 +31m TechCrunchN Nvidia launches new platform for reining in rogue AI agents
Times are when each article was published, or when we first saw it if the outlet gave no time.