From VentureBeat · 24 stories
Browser Security Gap Widens as Enterprise Work and AI Shift to Web Interfaces
Enterprise work increasingly occurs within web browsers, making them a primary target for cyberattacks. Traditional endpoint and network security architectures are insufficient for protecting browser sessions, especially with the rise of AI-assisted hacking and the use of third-party AI models.
Brex develops CrabTrap for securing AI agents by monitoring network activity
Brex CEO Pedro Franceschi announced CrabTrap, a new network-level security layer designed to secure AI agents like OpenClaw in production environments. This development addresses the challenge of safely deploying AI agents that can self-bootstrap and maintain their own codebases, which traditional security models struggled to control.
Thinking Machines Releases Inkling-Small AI Model, Nearing Predecessor's Performance at 1/4 Size
Thinking Machines released Inkling-Small, a new open-source AI language model that achieves performance comparable to its larger predecessor, Inkling, while being significantly smaller. This development matters because it offers enterprises a more efficient AI model with reduced compute requirements and inference costs, making advanced AI more accessible for organizations with limited GPU resources.
Mastercard Adapts Fraud Detection Systems for AI-Driven Transactions
Mastercard is re-evaluating its fraud detection systems to accommodate purchases made by bots, a shift from its traditional focus on preventing bot-driven fraudulent transactions. This change is necessary as AI increasingly facilitates legitimate transactions, requiring an update to Mastercard's risk framework.
Bright Machines Introduces Hybrid Robotic Cell to Integrate Human Operators into Automated Production
Bright Machines launched the Hybrid BRC (Bright Robotic Cell), an expansion of its Bright Factory platform, which allows human operators to perform assembly steps within a sensor-monitored robotic cell without disrupting the digital production record. This development addresses a challenge in high-stakes electronics manufacturing where manual intervention typically breaks the data thread, leading to low initial production yields for complex products like AI servers.
Visa open-sources vulnerability agentic harness after using Anthropic's Mythos to find bugs
Visa used Anthropic's Mythos AI model to identify vulnerabilities in its payment network infrastructure and subsequently open-sourced the "Visa Vulnerability Agentic Harness" used in the process. This initiative provides a reference implementation for other security teams to inspect, adapt, and extend their own vulnerability discovery efforts using AI.
New Ransomware Targets AI Model Weights, Lacks Exfiltration or Payment Mechanism
Sysdig's Threat Research Team documented a new ransomware, ENCFORGE, specifically designed to destroy AI model weights and training data, which was deployed after exploiting a Langflow vulnerability. This ransomware is unusual because it lacks network code for exfiltration or payment, focusing solely on data destruction, making recovery costly and complex for organizations.
Microsoft launches MAI-Image-2.5-Pro and MAI-Voice-2-Flash AI models
Microsoft AI released two new in-house models, MAI-Image-2.5-Pro for high-fidelity image generation and MAI-Voice-2-Flash for high-volume speech, into public preview. These models demonstrate Microsoft's commitment to powering its products with its own AI infrastructure, reducing reliance on external models like OpenAI's.
Inflection AI returns to consumer market with Pi Journeys and new research division
Inflection AI launched Inflection AI Labs and Pi Journeys, an experimental product focused on relational AI experiences. This move marks the company's return to the consumer market, emphasizing long-term user relationships over transactional interactions. The strategy aims to differentiate Inflection AI by addressing what it perceives as a flaw in current AI assistants.
Poolside launches Laguna S 2.1, an efficient 118B-parameter coding model
Poolside released Laguna S 2.1, a 118-billion-parameter open-weight coding model that activates 8 billion parameters per token. The model demonstrates competitive performance against larger models, addressing the growing demand for transparent open-weight AI solutions in the West.
Weka launches NeuralMesh 6 to enhance GPU memory via flash storage caching
Weka has unveiled its NeuralMesh 6 platform, which uses affordable flash storage to cache AI model tokens, reducing dependency on expensive GPU memory. This innovation could lower costs and speed up AI deployments, particularly benefiting organizations with high GPU utilization demands.
Expedia's AI chief proposes evals as a replacement for PRDs
Xavi Amatriain, Expedia's AI chief, introduced the concept of using evals instead of traditional PRDs for product development at VB Transform 2026. This shift emphasizes incorporating security and evaluation criteria from the outset, impacting how companies might approach product design and AI integration.
Brex develops CrabTrap for AI agent governance by analyzing real agent behavior
Brex created an internal platform named CrabTrap to enhance AI agent governance by analyzing actual agent behavior rather than relying solely on predefined rules. This shift utilizes an open-source proxy to intercept and evaluate network requests, addressing security challenges at the transport layer.
1Password introduces AI Spend Management to tackle new enterprise budgeting challenges
1Password has launched AI Spend and Consumption Management, offering real-time insights into AI service spending. This new capability addresses the rising budget pressures from AI usage as enterprises shift to consumption-based pricing models.
Study Reveals Enterprises Misjudging AI Model Failure Rates by 2.25x
A study of 67 AI models shows enterprises are miscalculating failure risks due to ignoring the 'co-failure ceiling.' This oversight results in ineffective multi-model strategies, leading to unnecessary complexity and costs.
Slackbot integrates with Salesforce for CRM and analytics via chat
Slack has launched an integration allowing Slackbot to access Salesforce CRM data and generate analytics charts through chat. This move enhances productivity for sales teams by enabling them to perform multiple tasks without switching applications.
Tencent releases Apache-licensed Hy3 model, outperforming rivals
Tencent has launched the Hy3 model with 295 billion parameters under an Apache 2.0 license, removing previous deployment restrictions. This change aims to enhance accessibility for enterprises, particularly in regions previously excluded from similar technologies.
AI-Powered Debate Explores America's Innovations for 250th Birthday
On America’s 250th birthday, 250 randomly selected participants used hyper-communication AI to deliberate on the nation's top innovations. This technology facilitates large-scale discussions, highlighting how AI can enhance collective intelligence in addressing significant topics.
Trunk Tools Reduces Document Review Time from 60 Days to 10 with Specialized AI Stack
Trunk Tools has developed a specialized AI architecture that reduces document review times from 60 days to 10 by addressing the challenges posed by general-purpose models in the construction industry. This tailored approach allows for more accurate automation and could serve as a model for other sectors facing similar data management issues.
Alibaba's SkillWeaver AI framework reduces tool token usage by 99%
Alibaba researchers developed SkillWeaver, a new framework that optimizes AI agent tool selection, reducing token consumption by 99%. This allows for greater accuracy in executing complex workflows by integrating a feedback loop to route tasks more efficiently.
Square launches ChatGPT and Claude ordering integration for restaurants
Square has launched a ChatGPT app and Claude plugin that allows consumers to order from restaurants directly within these AI platforms. This integration eliminates traditional marketplace fees while maintaining standard transaction fees, addressing high commission rates imposed by major delivery services.
Morgan Stanley reduces P&L reconciliation time by 50% with AI integration
Morgan Stanley has deployed an AI system, FIXR, to enhance profit and loss reconciliation, reducing the time required from six hours to two to three hours. The approach emphasizes human involvement over autonomy, increasing accuracy and efficiency in a critical workflow.
AI and human expertise must evolve together in IT and security fields
Organizations are increasingly using AI to automate tasks traditionally performed by junior analysts, which risks diminishing the apprenticeship opportunities crucial for developing skilled operators. Without intentional restructuring, companies may lose the deep analytical expertise needed to manage complex systems effectively in the future.
Runway ML turned an AI video generation bug into a feature for its Characters model
Runway ML addressed a bug in its AI video model, where AI-generated avatars drifted off-center, by implementing a front-end feature that worked around the problem rather than fixing the underlying issue. This approach was highlighted by Ryan Phillips, head of enterprise product at Runway ML, as a lesson in AI development and evaluation for companies building generative tools.