Anthropic Launches Staged Biology Research Access Program for Frontier AI Models

·2 views

Anthropic is creating a structured, phased access program allowing life-science researchers to use its advanced AI model Mythos 5, part of the company's biology-oriented Mythos and Fable line. An initial restricted cohort of biology researchers will gain access under modified cybersecurity safeguards, with broader access planned as those protections mature. The company frames this as a deliberate governance measure, separating frontier biology-capable models from general product availability rather than issuing a standard public release. Anthropic also maintains a separate External Researcher Access Program through which interested researchers can apply, though selection criteria and timelines have not been publicly detailed. The staged rollout reflects growing recognition that advanced AI capabilities in life sciences require more careful access controls than conventional software deployment.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How to Estimate GPU Memory Needs for AI Models Before You Deploy

A technical guide aimed at site reliability engineers explains how to systematically size GPU infrastructure for hosting AI models, rather than relying on guesswork. The method focuses on two main memory consumers: model weights, which are a fixed cost loaded once into VRAM, and the KV cache, which holds working memory for concurrent requests and directly determines throughput. Using Hugging Face's Qwen2.5-7B-Instruct-AWQ as a demonstration, the guide shows how quantization precision — fp16, int8, or 4-bit AWQ — dramatically changes a model's memory footprint, from roughly 15 GB down to under 6 GB for the same parameter count. The article notes that advertised weight sizes can differ from actual file sizes, and recommends verifying by summing model file sizes directly from the repository. The sizing process is described as a rule-of-thumb filter to narrow GPU family choices, with precise measurements recommended once the model is running via tools like vLLM.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

DSH Hub Offers Repository-First Directory to Vet DeepSeek Harness Plugins Safely

DSH Hub is a bilingual online directory designed to help developers evaluate DeepSeek Harness plugins and clients before installation by requiring every listed item to link to a public GitHub repository. The platform separates plugins from clients, recognizing that clients carry broader system access — including network listeners, runtimes, and extension bundles — and therefore warrant closer scrutiny. As of an August 18, 2026 catalog snapshot, the directory lists 19 plugins and 6 clients sourced from the official Harness repository, GitHub topic tags, and a community registry. The site's 'verified' label is limited in scope, confirming only that a repository was publicly reachable on the snapshot date, not that the code has been security-audited or that published artifacts match the source. DSH Hub positions itself as a faster starting point for research, not a substitute for personal technical review of permissions, licenses, and update paths before adding any extension to a working setup.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Anthropic Reports Claude Deployed in Live Research Workflows at Stanford and MIT

Anthropic published findings on January 15, 2026, describing Claude's use in active life sciences research at Stanford and MIT laboratories. The deployments leverage Claude for Life Sciences, a capabilities suite featuring improvements in Opus 4.5, access to over 60 scientific databases, and tools for genomics, proteomics, and cheminformatics. At Stanford's Biomni project, Claude assisted with molecular cloning design and large multi-source dataset analysis, with researchers reporting certain tasks completed in minutes rather than weeks. MIT's Cheeseman Lab developed a Claude-powered system called MozzareLLM to interpret large-scale gene knockout data, providing context-rich reasoning rather than bare answers. Anthropic frames these results as early-stage case studies of lab-scale use, with human scientific judgment and experimental validation remaining central to the research process.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Stale Lock File Silently Halted Automated Instagram System for Two Hours

A developer running a Claude Code-based autonomous Instagram content system lost two hours of output when a killed process left behind a lock file that no subsequent script run would override. The lock mechanism checked only file modification time, not whether the original process ID was still alive, causing every scheduled launch to exit cleanly with no content generated and no alerts fired. The silent failure went undetected precisely because the system was designed to run without human oversight, highlighting a key risk of unmonitored automation. The fix required adding a process-liveness check alongside the timestamp check, so a stale lock held by a dead PID is forcibly cleared. The incident underscores that robust automation must anticipate breakage and include self-repair logic, not just correct execution under ideal conditions.

0 comments Read more at DEV Community