For security work
AI for security work.No cyber-access application.
GLM-5.3 and an abliterated DeepSeek V4.1 Flash, ready on your Maple plan for code review, pentesting and incident response. Start today, not after an approval.
GLM-5.3 · DeepSeek V4.1 Flash Uncensored · Pro, Max and Team
The refusal wall
Why do AI models block defensive security work?
Because defensive work looks like attack work, and their guardrails can't tell an incident responder from an attacker. Hugging Face hit it: commercial models blocked its forensic analysis of attack payloads, so it ran GLM-5.2 on its own infrastructure. The programs that lift those blocks mean applying first. The incident write-up
Thomas Wolfco-founder, Hugging Face “When a frontier model is attacking you and moving laterally inside your infrastructure, defenders need wide access to near-frontier tools within hours or even minutes, rather than being pointed towards a closed-door, vetted application programme for model access.”
Models
Which LLMs work for cybersecurity?
Two open models in the picker on Pro, Max and Team. Rates are Maple Proxy API prices per million tokens.
- Pro
- Max
- Team
GLM 5.3glm-5-3
Deep review of one module, and agent work. Very capable in outside cyber tests.
Context262,144tokens
Input / output$3.60 / $11.50per 1M tokens
GuideDeepSeek V4.1 Flash Uncensoreddeepseek-v4-1-flash-1017-uncensored
Whole codebases in scoped passes, logs and payloads. Weights edited to reduce refusals.
Context1,048,576tokens
Input / output$0.65 / $1.45per 1M tokens
GuideAbliteration reduces refusals; it does not remove them. Check every finding before it reaches a patch. What Anthropic found about GLM-5.3
Private by proof
Can Maple see my code and findings?
What Maple sees
Nothing readable.
- Encrypted on your device.
- Decrypted only inside an attested enclave.
- Maple cannot see your chats.
Get started
How do I use these models with my security tools?
-
Step 1
Pick a plan
Pro, Max or Team. API access through Maple Proxy needs a paid plan; usage draws from your plan first.
-
Step 2
Get the proxy
It's built into the Maple desktop app as Local Proxy, for macOS, Windows and Linux. Or run it on its own with Docker or a release binary.
-
Step 3
Point your tool at it
Set the base URL and your Maple API key in OpenCode, Goose or any OpenAI-compatible client, then pick a model.
http://127.0.0.1:8080/v1glm-5-3 · deepseek-v4-1-flash-1017-uncensored
Further reading
How to use AI on your own code and advisories
- Guide AI code audit: check your code without waiting for cyber access Read the guide
- News PHP2Shell: the PHP vulnerability researchers found with AI Read the report
- Guide Abliterated DeepSeek V4.1 Flash for defensive security work Read the guide
- Guide GLM-5.3: capabilities, hardware and access options Read the guide
- News Cloudflare Workers AI KV cache leak: what happened, and what to check Read the report
- News Anthropic's GLM-5.3 cyber tests: results, limits, and NIST's findings Read the analysis
Questions about AI for security work
Do I need to apply for cyber access?
No. Both models are in the picker on Pro, Max and Team. There's no application and no wait.
What is an abliterated model?
A model whose refusal behavior is reduced by editing its weights. It can still decline, and you still review its output.
Which Maple plans include these models?
Pro, Max and Team. In the picker they are GLM-5.3 and DeepSeek V4.1 Flash Uncensored.
Can Maple see my prompts and code?
No. Maple cannot see your chats. You can run the live attestation check yourself on /proof.
Does it work with my existing security tools?
Yes, through Maple Proxy, which is OpenAI-compatible. Change the base URL and keep your tools.