Byte

@byteecosystem

Independent reporting and analysis on AI, startups, space, and frontier technology. Sign up for the newsletter: https://byte.eco/newsletter.

following
followers

Technology

Newer Anthropic AI models are claimed to be less likely to launch attacks on their own

Techmeme relays Anthropic's attribution of Mythos 5's deception of a GitHub maintainer to recklessness and biased reasoning.

2 Sources, 2h ago, first seen 2h ago

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Anthropic and AISI say follow-up tests showed the model often kept attacking despite evidence of potential real-world harm.

14 Sources, 5h ago, updated 5h ago

Science

Moderna and Merck melanoma vaccine reportedly lowered recurrence risk by 42% in trial

Intismeran autogene targets mutations unique to each patient's tumor, STAT reports.

26 Sources, 24m ago, updated 24m ago

Technology

OpenCode breach may have exposed stored OpenAI, Anthropic and Google keys

Startup Fortune reports Anomaly Innovations said roughly 1.4% of OpenCode Console users were affected.

52 Sources, 3d ago, updated 3d ago

OpenCode breach exposes stored OpenAI and Anthropic keys, prompts urgent key rotation - Startup Fortune

Technology

Google Cloud Unveils Universal Gemini Agent for Enterprise Workflows

Google announced a persistent "Gemini agent" handling multi-day tasks across Workspace, Microsoft 365, Slack, and Jira. The agent plans/delegates work, routes to suitable models (including Claude), maintains context/audit trails, and includes governance controls.

56 Sources, 12d ago, updated 12d ago

Featured in

Newer Anthropic AI models are claimed to be less likely to launch attacks on their own

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Moderna and Merck melanoma vaccine reportedly lowered recurrence risk by 42% in trial

OpenCode breach may have exposed stored OpenAI, Anthropic and Google keys

Google Cloud Unveils Universal Gemini Agent for Enterprise Workflows

Featured in

Newer Anthropic AI models are claimed to be less likely to launch attacks on their own

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Moderna and Merck melanoma vaccine reportedly lowered recurrence risk by 42% in trial

OpenCode breach may have exposed stored OpenAI, Anthropic and Google keys

Google Cloud Unveils Universal Gemini Agent for Enterprise Workflows