Airlock Digital has launched Agentic AI Control & Governance, a tool to help organizations govern AI agents and ensure they operate within organizational policy. This tool provides visibility into AI agent behavior, command- and session-level visibility, and real-time policy evaluation.
Researchers have discovered knowledge holes in unlearned multimodal large language models (MLLMs), which can lead to severe degradation on benign adjacent inputs. To address this issue, researchers have proposed a new benchmark and a method called Selective Protection with Anchored Regularization.
New York state agencies are required to report their AI use, but only 13 out of 100 agencies publicly reported using AI in a way that could directly impact the public. The European Union has introduced new rules requiring AI systems to be labeled and transparent, with fines of up to $17.2 million for non-compliance.
Google has deleted three AI workflows from its ADK repository after a security researcher discovered a vulnerability that could trigger a privileged agent and reach code execution on a CI runner. AI-generated reports are flooding the CVE pipeline with fake vulnerabilities, leading to false positives and decreased system reliability and security.
Moores Lab AI has launched SpecAgent, a product for full-lifecycle silicon engineering. A German court has ruled that AI music platform Suno infringed copyright by training its music generation model on protected songs without permission.
Elon Musk has warned of AI's potential impact on the economy, describing it as a 'supersonic tsunami.' He emphasized that while AI is expected to drive growth, investors are increasingly focused on the costs of building out AI infrastructure.
Key Takeaways
* Airlock Digital launches Agentic AI Control & Governance to help organizations govern AI agents.* Researchers discover knowledge holes in unlearned multimodal large language models (MLLMs).
* New York state agencies struggle to track AI use, with only 13 out of 100 agencies publicly reporting.
* Europe introduces AI labeling and transparency rules with fines of up to $17.2 million.
* Google deletes AI workflows after security issue discovered.
* AI-generated reports flood CVE pipeline with fake vulnerabilities.
* Moores Lab AI launches SpecAgent for full-lifecycle silicon engineering.
* German court rules Suno's AI training infringed copyright.
* Elon Musk warns of AI's potential impact on the economy.
* Investors focus on costs of building out AI infrastructure.
Airlock Digital Launches AI Control and Governance Tool
Airlock Digital has introduced Agentic AI Control & Governance, a new tool to extend preventative endpoint security. This tool helps organizations understand AI agent behavior, define trusted operating boundaries, and govern AI activity at the endpoint. It also provides visibility into AI agent behavior, command- and session-level visibility, and real-time policy evaluation. The goal is to help security and IT teams govern AI agents and ensure they operate within organizational policy.
Airlock Digital Unveils Agentic AI Control & Governance
Airlock Digital has launched Agentic AI Control & Governance, a new tool to help organizations govern AI agents and ensure they operate within organizational policy. This tool provides visibility into AI agent behavior, command- and session-level visibility, and real-time policy evaluation. It also helps security and IT teams understand AI agent behavior, define trusted operating boundaries, and govern AI activity at the endpoint.
Researchers Explore Knowledge Holes in Unlearned Multimodal Large Language Models
Researchers have discovered knowledge holes in unlearned multimodal large language models (MLLMs). These knowledge holes can lead to severe degradation on benign adjacent inputs. To address this issue, researchers have proposed a new benchmark to capture unintended degradation on benign inputs sharing generic patterns with the forget set. They also propose a new method, Selective Protection with Anchored Regularization, to bridge the gap.
New York Struggles to Track AI Use in State Agencies
New York state agencies are required to report their AI use, but only 13 out of 100 agencies publicly reported using AI in a way that could directly impact the public. The disclosures were part of an inventory under the state's 2024 policy governing AI use. However, several high-profile AI governance challenges arise from the lack of transparency and inconsistent reporting.
Europe Introduces AI Labeling and Transparency Rules
The European Union has introduced new rules requiring AI systems to be labeled and transparent. The rules aim to help users identify AI-generated content and ensure that AI systems are trustworthy and accountable. Companies that fail to comply with these rules risk facing fines of up to $17.2 million.
Google Deletes AI Workflows After Security Issue Discovered
Google has deleted three AI workflows from its ADK repository after a security researcher discovered a vulnerability that could trigger a privileged agent and reach code execution on a CI runner. The issue was identified and reported to Google, which removed the workflows and confirmed the issue was fixed.
AI-Generated Reports Flood CVE Pipeline with Fake Vulnerabilities
The National Institute of Standards and Technology (NIST) is struggling to keep up with its backlog of Common Vulnerabilities and Exposures (CVEs), and AI-generated reports are flooding the pipeline with fake vulnerabilities. This can lead to false positives and a decrease in system reliability and security.
Moores Lab AI Launches SpecAgent for Full-Lifecycle Silicon Engineering
Moores Lab AI has launched SpecAgent, a new product in its MooreIP platform for full-lifecycle silicon engineering. SpecAgent helps teams automate the silicon development lifecycle, from specification through signoff. It ingests requirements from existing documentation and communication channels and organizes design intent into consistent, traceable requirements.
German Court Rules Suno's AI Training Infringed Copyright
A German court has ruled that AI music platform Suno infringed copyright by training its music generation model on protected songs without permission. The court ordered Suno to pay damages and disclose revenue related to the infringement. This ruling is seen as a landmark case for creators, highlighting that AI models built on stolen intellectual property have no protection under the law.
Elon Musk Warns of AI's Potential Impact
Elon Musk has described AI as a 'supersonic tsunami' and warned of its potential impact on the economy. While AI is expected to drive growth, investors are increasingly focused on the costs of building out AI infrastructure. The next phase of the AI trade may hinge on investor tolerance for infrastructure spending, and companies may need to generate returns quickly to justify the unprecedented wave of spending.
Sources
- Airlock Digital Unveils Agentic AI Control & Governance to Extend Preventative Endpoint Security
- Airlock Digital Unveils Agentic AI Control & Governance to Extend Preventative Endpoint Security
- Exploring and Bridging Knowledge Holes in Unlearned Multimodal Large Language Models
- The First Step to Governing AI? Finding It.
- Europe’s AI labeling and transparency rules are now in effect
- Google Deletes 3 ADK AI Workflows After Malicious GitHub Issue Could Trigger Privileged Agent
- AI slop pollutes the CVE pipeline with fake vulns
- Moores Lab AI Launches SpecAgent, the Latest Product in its MooreIP Platform for Full-Lifecycle Silicon
- German court says Suno’s AI training infringed copyright in landmark ruling
- Elon Musk Says AI Is A 'Supersonic Tsunami.' Wall Street Is Worried About The Bill
Comments
Please log in to post a comment.