A busy day of AI developments spanning promising scientific breakthroughs and concerning safety vulnerabilities, alongside a security escalation in enterprise software and cloud integrations that make application development easier.
AI Between Scientific Progress and Safety Tests
Designing enzymes to address neurological diseases: Researchers from David Baker’s team at the University of Washington developed AI-designed proteases to target proteins associated with Alzheimer’s disease and frontotemporal dementia. The models showed measurable activity and improved through mutations, but they remain less efficient than natural enzymes and far from therapeutic use.
Image-generation guardrails under pressure: A test by White Circle revealed that it was possible to bypass the filters of DALL·E 3, Nano Banana 2, and GPT-Image-2 to produce explicit images. The findings highlight the need for independent, mandatory safety testing, particularly given the varying robustness of newer releases.
AI Governance and Enterprise Applications
Anthropic expands in Singapore: The company opened its fifth office in Asia-Pacific and signed a memorandum of cooperation with the Singaporean government on model safety, security, evaluation, and responsible adoption, in a market where government, financial, and technology organizations use Claude services.
Security for coding agents: Chainguard launched the Harness platform to integrate exploitable-vulnerability discovery with isolated microVM environments and CI checks into coding-agent workflows. This comes amid an indication that 66% of companies with more than $1 billion in revenue may sometimes leave high- or critical-severity vulnerabilities unpatched.
Enterprise Security and Cloud Infrastructure
SAP patches critical vulnerabilities: SAP issued 22 security notes and updates on its October Patch Day, including four critical vulnerabilities with scores of up to 10 out of 10. Some of the flaws could enable code execution or internal operations and full control of affected systems.
S3 joins the Vercel ecosystem: Amazon S3, including S3 Tables and S3 Vectors, is now available through the Vercel Marketplace and v0. The integration enables the creation of storage resources and their connection to applications without manually configuring static access keys, with support for public storage, Apache Iceberg analytics, and vector search for AI applications.