Offense changelog

Product updates, improvements, and fixes.

Red team, without the setup friction #

Automatic swarm planning, live progress, and clearer reports.

15 Nov 2026

Improvement
Fix

Launch with a target, not a prompt

Swarm agents now choose their own focus areas when you leave steering blank. Add attack vectors, focus areas, or security context only when you want to guide the work. Test identities are optional for public-surface coverage; provide them when you need authenticated testing.

Follow the work as it happens

Reports show queued, running, completed, and stopped agents, with live updates and a combined view of findings. You can clear and replace the agent count while setting up a launch.

Get to the results faster

Signing in now opens Red team. Reports put findings and agent outcomes first, prioritize higher-severity findings, and keep evidence, logs, and usage details expandable.

Smoother billing and model access

The checkout link no longer lands on a missing page. First-time basic KYC for models that require it is available directly in the launch flow.

Open Red team · Read the docs

A simpler start for your coding agent #

Start a sandbox from your agent and give it work when you are ready.

14 Nov 2026

Improvement

Connect, create, then get to work

The agent setup page now leads with a copyable prompt for connecting to Offense and creating a sandbox. A target and test identities are not required just to create the environment.

Sandbox tools in MCP

The updated MCP package exposes sandbox creation, status, and stop tools. Agents can show you the sandbox link and check readiness before continuing. Creation retries retain their request identity to avoid accidentally creating a second sandbox.

More useful local project runs

One-shot CLI runs include nonignored, untracked project files, so newly created files can be part of the work before they are committed.

Connect your agent

Stronger verification and coordinated sweeps #

Better evidence for fixes and more structured ways to investigate dependencies.

13 Nov 2026

Feature
Improvement

Check whether a fix addresses the original finding

Patch-proof results compare the original finding with verification observations. An inconclusive retest stays inconclusive instead of being treated as proof that the issue is fixed.

Plan larger investigations in bounded batches

The API can plan and execute dependency and compromise sweeps in bounded batches. Assignments retain a defined scope and budget, making larger investigations easier to track.

Carry context between code and live assessments

Routes discovered by a completed code scan can be carried into a live assessment. Externally reported findings can also enter the workflow as candidates for verification.

More resilient report access

Findings remain readable when database rollout is one migration behind the application.

Review findings

Keep security work moving after the first scan #

Recurring assessments, outcome notifications, and follow-up workflows.

12 Nov 2026

Feature

Run assessments on a schedule

Security loops let you schedule repeat assessments against a configured target, with run budgets and optional testing windows.

Bring results into your workflow

Connected Slack workspaces can receive terminal assessment outcomes. GitHub follow-up workflows connect findings to remediation pull requests, and supported AWS connections can collect log evidence for an investigation.

More control over sandbox work

Sandbox APIs gained command execution, per-job network policy, and structured result handling, giving clients a clearer way to coordinate work and consume its output.

Manage security loops · Manage connections

Swarms, repository scans, and connected workflows #

Multiple agents, one target, and a combined report.

11 Nov 2026

Feature

Launch a coordinated swarm

Run multiple agents against one target, each with its own assignment. The combined swarm report brings their findings together and merges duplicates so you can review results in one place.

Scan connected repositories

GitHub repository connections support code and dependency scans. Install the GitHub App from Connections to connect the repositories you want to assess.

Connect your existing infrastructure

Connection setup expanded to Cloudflare, AWS, and Vercel, with domain import for supported connections. Workspace settings let you choose supported model and provider preferences.

Work from your terminal

The Offense CLI and scoped API keys provide a terminal workflow for signing in and managing work. CLI 0.1.1 is available through the app’s package download.

Explore Offense