Superpower Daily
OpenAI Releases GPT-6 Astra With Its First Advanced Cyber Safeg...Microsoft Releases MAI-Transcribe-2 at 10 Cents an HourFigure Signs $3.5B Nscale Deal for Up to 100,000 GPUs for Human...Trump Order Opens Path to Restrict Foreign Power Gear for AI Da...Cohere Publishes Agent-Tool Dataset, Finds 419 Occupations Have...Google Launches Voice AI in Gmail, Docs and Keep, With Business...OpenAI Releases GPT-6 Astra With Its First Advanced Cyber Safeg...Microsoft Releases MAI-Transcribe-2 at 10 Cents an HourFigure Signs $3.5B Nscale Deal for Up to 100,000 GPUs for Human...Trump Order Opens Path to Restrict Foreign Power Gear for AI Da...Cohere Publishes Agent-Tool Dataset, Finds 419 Occupations Have...Google Launches Voice AI in Gmail, Docs and Keep, With Business...

Full archive

Latest AI News

Search reporting on AI markets, models, products, companies, policy, and culture.

Category

More filters
Clear all
Showing 1-18 of 33Results matching "Agent harness"

Today

1 story

Yesterday

3 stories
LaunchTools

Google Launches Fairwind to Give Selected Defenders AI Tools for Finding and Fixing Flaws

The program draws a line between broadly available code-security tooling and restricted access to Gemini 3.8 Flash Cyber, reflecting the dual-use risk of AI that can identify and repair software weaknesses. Google has launched the Fairwind Program, a limited-access channel for governments and trusted partners to use AI systems that can find, verify and fix software vulnerabilities.

3 min readRead story

Earlier this week

5 stories
LaunchProducts

CrowdStrike’s SafeMind Pits Red and Blue AI Agents in an NVIDIA Infrastructure Twin

The new Falcon system combines CrowdStrike’s security data and agent harnesses with NVIDIA’s open Nemotron models. Early tests favor the optimized open pipeline on detection quality, but cover only one attack scenario family. CrowdStrike has introduced SafeMind inside its Falcon platform, pairing its cybersecurity models and agent harnesses with NVIDIA Nemotron open models.

3 min readRead story
BenchmarkModels

Grok 4.6 Clears 50% on Biosecurity Refusal and Routine Biology Tasks

The reported result rewards a difficult balance: recognizing hazards hidden inside ordinary-looking research requests without shutting down legitimate biological work. Its surveillance score shows that refusal performance is not the whole biosecurity test. The result puts the model’s ability to distinguish intent at the center of a biosecurity contest where excessive blocking can also impair useful work.

3 min readRead story
BenchmarkModels

Qwen3.8-Max Leads Open-Weight Commerce Agent Bench by Two Passes, With Audit Limits

The benchmark checks whether agents leave commerce software and files in the required state, rather than simply narrate a workflow. But configuration choices and unavailable immutable result bundles limit what the narrow ranking can settle. Qwen3.8-Max holds a two-pass lead over DeepSeek V4 Pro in the published open-weight Commerce Agent Bench results, with 156 passes across three agent harnesses.

2 min readRead story
Open releaseModels

DeepSeek Releases 305B Vision Weights Under MIT License, but Local Use Needs Serious Hardware

The release gives developers the model files and serving references needed to run or adapt DeepSeek-V4-Flash-Vision-Exp independently. Its 168GB size means that freedom is aimed more at infrastructure-equipped teams than typical individual users. The shift gives developers more control over a model that accepts images, but the 168GB checkpoint sets a high infrastructure threshold for anyone seeking to run it locally.

3 min readRead story
LaunchProducts

Modellix Adds 28+ Models, Including $0 Options, to DeepSeek Harness Beta

The beta gives users of the rapidly adopted coding agent one model-selection layer and web-tool routing, while its free-call pitch still requires a Modellix account and funded balance to create an API key. Aurora Mobile’s Modellix has released dsh-modellix, a beta plugin that brings its LLM gateway into DeepSeek Harness ’s native model selector. The plugin’s LLM, Design and Web functions can be toggled independently.

3 min readRead story

Last week

9 stories
Platform shiftProducts

OpenAI Maps a Third Era of Persistent AI Coworkers, With Access and Reliability Still in the Way

The proposed shift would turn AI from a prompt-by-prompt helper into a longer-running collaborator, but its usefulness depends on shared context, workplace-system access, and dependable controls. OpenAI ’s next workplace interface, in product lead Tara Seshan’s telling, is not a better chatbot or a one-person agent. That framing changes the unit of work.

3 min readRead story
BenchmarkTools

Perplexity Search API Sweeps Artificial Analysis Test, but Medium Context Wins

The result gives developers a concrete retrieval-quality lead to test, while showing that sending an agent more page text is not automatically the best route to better answers. Perplexity ’s Search API occupied the top three positions in Artificial Analysis’ Search Index, with its medium-context setting scoring 80. Search APIs give an AI agent web results and extracted page material to reason over.

3 min readRead story
BenchmarkModels

Meta’s EvoHarness-RL Takes Qwen3-8B to 96.9% on ALFWorld, Near Claude Opus 4.5

The result is a strong benchmark showing for learned state management, not yet evidence that the framework improves the long-running enterprise workflows it is designed to target. Meta AI and the University of Illinois Urbana–Champaign say their EvoHarness-RL framework let Qwen3-8B reach 96.9% average success on ALFWorld, edging Claude Opus 4.5’s reported 96.4% score. Source: venturebeat.com.

3 min readRead story
Open releaseTools

Visa’s VVAH Adds AI Patch Writing Across 11 Stages, Leaving Merge Decisions to Humans

The open-source harness now moves from detection through candidate remediation and adversarial validation. Its default can alter a repository’s working copy, while build, test, review, and merge controls remain outside the tool. The consequential boundary is that VVAH edits source files by default, but Visa says it does not merge changes or replace an organization’s normal build, test, and review flow.

3 min readRead story
LaunchProducts

Harness Launches Agent-Ready Repository and AI Review for High-Volume Code Changes

The products pair high-volume repository operations with inherited permissions and merge-time checks, but Harness still places the final decision to ship with people. Harness has launched an agent-ready code repository and AI code-review product built for a delivery problem that emerges after code generation: agents can create changes faster than conventional pull-request workflows can absorb.

2 min readRead story
LaunchInfrastructure

AWS Publishes Two AgentCore Paths to Query Cross-Account Knowledge Bases

The useful constraint is not simply access across AWS accounts: generated answers require an assumed role, while many workloads may not need an agent at all. AWS has published two generally available AgentCore patterns for agents that need answers from a Bedrock Knowledge Base held in a separate AWS account. In AWS’s example flow, a user submits a natural-language question through an API or Streamlit interface.

3 min readRead story
LaunchModels

DeepSeek V4 Pro Lands on Fireworks With $2.50 CyberGym Solves, but Kimi K3 Scores Higher

The provider’s results frame security-agent selection as a three-way choice among raw solve rate, cost per completed task, and whether a model will reliably execute the required tool workflow. Fireworks AI has made DeepSeek-V4-Pro-0813 available on its platform, pairing the release with security-benchmark results that favor the model on cost and tool reliability rather than outright solve rate.

3 min readRead story
15 more next