OpenAI is making significant advancements in AI security with its new model, Astra, which has shown critical cybersecurity capabilities. The model is being evaluated under OpenAI's Preparedness Framework, which assesses potential risks and capabilities. Astra's preliminary results indicate strong performance, but OpenAI is taking steps to ensure safe development and deployment.
Anthropic is also prioritizing AI security, refining its Fable 5 model's safety protocols to reduce 'false positives' and broaden its utility for everyday health and educational inquiries. The updates aim to enable Fable 5 to assist with a wider range of biology tasks, including interpreting lab results and understanding symptoms.
Meanwhile, OpenAI's ChatGPT is getting updates to improve everyday conversations and expand access for Free users. The company is also working with Anthropic to address concerns around AI safety and security. Other developments include DXC partnering with Primary on AI security and Nevari launching Kassper.AI, a managed outbound service that books qualified meetings on behalf of B2B teams and founders raising investment.
Key Takeaways
- OpenAI is developing Astra, a new AI model with critical cybersecurity capabilities.
- Astra is being evaluated under OpenAI's Preparedness Framework to assess potential risks and capabilities.
- Anthropic is refining Fable 5's safety protocols to reduce 'false positives' and broaden its utility for everyday health and educational inquiries.
- OpenAI and Anthropic are prioritizing AI security and addressing concerns around AI safety and security.
- OpenAI's ChatGPT is getting updates to improve everyday conversations and expand access for Free users.
- DXC has partnered with Primary on AI security to help enterprises and government agencies govern AI agent and enterprise AI application access.
- Nevari has launched Kassper.AI, a managed outbound service that books qualified meetings on behalf of B2B teams and founders raising investment.
- Cambricon Technologies reported a 108% surge in first-half revenue, reaching 6 billion yuan.
- India is experiencing significant AI-driven job growth, with 83,100 hires compared to 31,921 layoffs.
- Multimodal LLMs can improve evidence-centric reasoning, but a multimodal integration paradox can degrade constrained resource allocation.
OpenAI develops Astra model with critical cybersecurity capabilities
OpenAI is working on a new AI model called Astra that may have critical cybersecurity capabilities. The model is designed to improve the security of AI systems and prevent breaches. Astra is still in development, but OpenAI says it could be used to improve the security of AI systems and prevent similar breaches in the future. The announcement comes amid a rash of AI model hacks in recent months. OpenAI's Astra model is expected to provide enhanced security features such as encryption and secure data storage.
OpenAI shares details of Astra model
OpenAI shared details of its upcoming Astra model, which has shown significant advancements in agentic coding and cybersecurity. The model is being evaluated under OpenAI's Preparedness Framework, which assesses the potential risks and capabilities of its AI models. Astra's preliminary results indicate strong enough performance that OpenAI cannot rule out Critical capability level at this time. The company is taking steps to ensure the safe development and deployment of Astra.
OpenAI flags critical cyber risks in Astra model
OpenAI announced that its upcoming model, Astra, is demonstrating significant advancements in agentic coding and cybersecurity. Internal evaluations suggest the model may possess capabilities that warrant classification under the 'Critical' tier of OpenAI's Preparedness Framework. Astra's preliminary results indicate a potential leap to 'Critical' capabilities, a development OpenAI is sharing transparently with the public and the broader safety and security communities.
OpenAI slows down Astra model development
OpenAI has suspended work on some aspects of its upcoming model Astra after an internal review found it had made significant advancements in agentic coding and cybersecurity. OpenAI says it is taking action, including enacting stricter security controls and pausing internal activities involving Astra that don't meet these beefed-up guardrails. The company is working with relevant government agencies and select AI safety organizations to test the capabilities for this model.
OpenAI and Anthropic prioritize AI security
OpenAI and Anthropic are prioritizing AI security, with OpenAI pledging to add Astra security and Anthropic loosening the leash on Fable. The companies are working to address concerns around AI safety and security, with OpenAI implementing stricter security controls for higher-capability models and Anthropic refining the safety protocols for its Fable 5 model.
Anthropic refines Fable 5 biology AI
Anthropic is refining the safety protocols for its Fable 5 AI model to broaden its utility for everyday health and educational inquiries. The updates aim to reduce 'false positives' from Fable 5's safety classifiers, enabling more support for tasks such as interpreting lab results and understanding symptoms. The company views biology and medicine as prime areas for AI's positive impact.
Anthropic improves Fable 5 safeguards
Anthropic is making updates to Fable 5's biology safeguards to substantially reduce false positives. Fable 5 users will experience many fewer fallbacks on everyday health and educational questions. The updates will enable Fable 5 to assist with a wider range of biology tasks, including interpreting lab results and understanding symptoms.
Multimodal LLMs aid decision-making
A new benchmark, C-SUITEBENCH, evaluates the decision-making abilities of multimodal LLMs in executive business decisions. The benchmark finds that multimodal inputs consistently improve evidence-centric reasoning, with the largest gains in risk forecasting and board-facing justification. However, the study also uncovers a multimodal integration paradox, where adding visual business information degrades constrained resource allocation.
Bayesian Expected Uncertainty Reduction model
The Bayesian Expected Uncertainty Reduction (B-EUR) model formalizes the value of trying a candidate design action as its expected reduction of epistemic uncertainty about action-outcome relations. The model addresses one part of the Uncertainty Driven Action (UDA) model's open question concerning how changes in uncertainty perception determine action selection.
AI tools aid back-to-school preparations
Parents can use AI tools like ChatGPT to create stress-free schedules and meal plans tailored to their needs. By using specific prompts, parents can get tailored solutions in seconds. AI tools can help manage busy schedules and streamline daily routines.
DXC partners with Primary on AI security
DXC has partnered with security startup Primary to become the exclusive managed services provider for Primary's AI-native Zero Trust platform. The joint offering is designed to help enterprises and government agencies govern how AI agents and enterprise AI applications access data, identities, and business systems.
AI-driven job growth in India
A report by Nomura indicates that India is experiencing significant AI-driven job growth, with 83,100 hires compared to 31,921 layoffs. Most firing cases are support teams replaced by chatbots, while most hiring cases are of IT-services graduates that firms explicitly credit to AI demand.
Cambricon sees revenue surge
Cambricon Technologies reported a 108% surge in first-half revenue, reaching 6 billion yuan. The Chinese AI chip giant saw profits jump 122.6% year on year to 2.3 billion yuan. The revenue surge is attributed to increasing demand for AI computing power in China.
Nevari launches Kassper.AI
Nevari has launched Kassper.AI, a managed outbound service that books qualified meetings on behalf of B2B teams and founders raising investment. Kassper pairs a proprietary AI agent stack with a human Nevari operator, and is built to move from sign-off to a first booked meeting within two weeks.
OpenAI improves ChatGPT and GPT-5.6 Sol
OpenAI is introducing updates to ChatGPT that improve everyday conversations while expanding access for Free users. The company is updating GPT-5.6 Sol in ChatGPT to be more reliable with facts and provide more focused answers. A new slider lets users choose how much thought ChatGPT puts into each response.
Sources
- OpenAI says its upcoming Astra model may have 'critical' cybersecurity capabilities, amid rash of AI model hacks
- Responding to the next frontier of critical cyber capabilities
- OpenAI Flags Critical Cyber Risks in Astra Model
- OpenAI says it slowed Astra model development over security concerns
- OpenAI pledges to add Astra security as Anthropic loosens Fable's leash
- Anthropic Tweaks Fable 5 Biology AI
- Improving Fable 5 Safeguards
- Seeing Is Not Deciding: Can Multimodal LLMs Act as Effective CEOs?
- Bayesian Expected Uncertainty Reduction (B-EUR) Model: A Computational Account of What Makes Design Options Worth Trying
- How AI can help parents manage back-to-school schedules
- DXC Becomes Exclusive Managed Services Provider for Primary’s AI Security Platform
- AI-related hiring amounted to 83100, 31,921 layoffs & attrition in India: Report; economists say ‘most firing cases….’
- Cambricon posts 108% surge in revenue amid China’s massive AI chip drive
- Nevari Launches Kassper.AI: The Sales Rep That Books Meetings While You Sleep
- Improving GPT-5.6 Sol in ChatGPT—and expanding access for free users
Comments
Please log in to post a comment.