Before you copy even a single line of log data into an AI tool, you should know where that data ends up. This is exactly the question every good security professional asks first, and it often matters more than the quality of the tool's answers. This tutorial walks you through the most important options: ChatGPT, Claude, local models with Ollama, and the enterprise variants. For each one, you'll learn how training with your content works, how long data is stored, and which compliance aspects are relevant. I'll also show you how to turn off training with your inputs in ChatGPT.
The video is only loaded from YouTube when you play it – data (including your IP address) is then transmitted to YouTube/Google. More in the privacy policy
Key Takeaways
Key Takeaways
- With the free version of ChatGPT, your inputs are used for training by default. You have to disable the option yourself in the data controls.
- With up to 200,000 tokens, Claude offers the larger context window and is particularly suitable for long log files and structured reports. With paid accounts, according to the video, Anthropic does not train on your content by default.
- Ollama runs open-source models such as Llama 3.2 locally on your machine. The data does not leave your device, but performance is lower than with large cloud models.
- Enterprise solutions such as Azure OpenAI or Claude for Business offer contractually guaranteed privacy rules, audit logs, and SLAs.
- Before using a tool, you should always check your company's policies.
Step-by-Step Guide
Step 1: Understanding the Default Settings of the ChatGPT Free Version
Before you start working with the tool, you should know what the default setting is. With the free version of ChatGPT, your inputs are used by default to train the models. This means that prompts containing internal information could indirectly flow into future model versions. Unless you actively intervene, this setting therefore applies automatically. The Plus version (around 20 US dollars per month) offers more queries and access to newer models, but you also have to check the training setting there.
Step 2: Finding the Data Controls in ChatGPT
To turn off training, open your account in ChatGPT and select the settings there. In the menu you will find the item Data Controls. The interface is available in German, so the labels are easy to find. Take a moment here to look through the options calmly, because the data controls are the central place for all settings related to your data.
Step 3: Disabling Training with Your Content
Within the data controls, you will find the Model Improvement switch, i.e. the option to improve the model for everyone. The text explains that by enabling it, you allow your content to be used for training. When the switch is active, your data is used for training purposes. Turn it off so that your inputs no longer flow into training. This step is one of the first measures you should take with every new installation.
Step 4: Assessing Claude and Its Large Context Window
Claude from Anthropic is particularly well suited to complex analyses. The decisive advantage is the context window: while ChatGPT can process around 128,000 tokens according to the video, Claude handles up to 200,000 tokens. In practice, this lets you analyze significantly more log files at once. The numbers have shifted further since the latest model generations, but the direction remains the same. For long documents and extensive log analyses, Claude is therefore often the better choice.
Step 5: Checking Privacy in the Claude Subscriptions
Claude is available in different tiers: a free version with limited usage, the Pro version, and Business versions with enterprise features such as SSO and admin controls. An important point for you: according to the video, Anthropic does not train its models on the content of paid accounts by default. This is a plus, so you don't have to actively disable anything here. Still, check the current terms of use, because policies change. For logs containing personal data, you should additionally clarify the data protection law permissibility on your own.
Step 6: Installing Ollama and Starting a Local Model
If your data must not leave the company under any circumstances, Ollama is the right solution. With Ollama, you run open-source models directly on your own computer. With the command ollama run llama3.2, for example, you start Meta's Llama 3.2 model, which is free to use. After the download, it runs entirely locally and also works without an internet connection. The costs are limited to your hardware.
A practical tip: In the video, the network traffic was checked with trackers, and no data stream went outward. Nevertheless, verify this yourself in your own environment, for example with a firewall rule or a network monitor. A claim is not the same as proof.
Step 7: Classify enterprise variants for companies
If your company already uses Microsoft Azure, Azure OpenAI is a good fit. There, the same GPT models run in your own Azure environment, with all the compliance features your company needs. Alternatively, there is Claude for Business from Anthropic or Amazon Bedrock, where you can use various models via AWS. These solutions offer contractually guaranteed data protection, audit logs, compliance reports, availability SLAs, and integration into existing security infrastructures. It is best to talk to your IT department, because such solutions are often already available without you knowing about it.
Step 8: Test the same scenario in all tools
To experience the differences in practice, you ask all tools the same question. In the video, it was: How do I recognize Windows event logs of a brute-force attack, and which event IDs are relevant? ChatGPT delivered a quick and concise explanation. Claude answered in more detail, more structured, and with concrete PowerShell queries. Ollama took a little longer but delivered a usable answer that stayed local. The quality was okay to good for all tools. The real differences lie in the surrounding aspects: speed, context size, and above all where your data ends up.
Step 9: Apply the rule of thumb for tool selection
A simple decision aid helps you in everyday work. For quick questions and general research, you use ChatGPT with web search. For long logs and complex analyses, Claude is the right choice, since it has a larger context window and prepares data in a structured way. As soon as sensitive company data is involved, you turn to Ollama or to an enterprise solution such as Azure OpenAI or Claude for Business. With specific compliance requirements, the respective requirement decides, and then enterprise solutions or local operation are usually the safe option.
Step 10: Check the company policy before any use
The most important step comes last, and you should never skip it: Check your company's policies before you use any tool. Many companies already have guidelines on which AI tools are permitted and which data may be entered. Anyone who acts on their own here risks not only data protection problems but also employment law consequences. When in doubt, ask before you upload sensitive logs.
Summary ??? AI Tools for Security Professionals: Data Protection, Training, and Tool Selection
Choosing the right AI tool is first decided by the question of where your data ends up. ChatGPT-Free trains on your inputs by default, so you have to turn off the option in the data controls yourself. Claude scores with a large context window for long logs and, according to the video, does not train on your content for paid accounts. Ollama offers the highest data security since models run locally, but you are limited in performance. Enterprise solutions such as Azure OpenAI or Claude for Business provide contractually secured compliance features. Use the rule of thumb: quick questions with ChatGPT, long logs with Claude, sensitive data locally or via enterprise solutions, and always check your company's policies beforehand.
FAQ
Are my inputs used for training in the free ChatGPT version?Yes, in the Free version, data is used for training by default until you deactivate the option in the data controls.
Where do I find the setting to turn off training in ChatGPT?You open your account, select the settings, and there choose the Data Controls item. There you deactivate model improvement.
Does Anthropic train its Claude models on my data?According to the video, Anthropic does not train on the content by default for paid accounts. Check the current terms of use, as these may change.
What is the advantage of Claude with long log files?According to the video, Claude processes up to 200,000 tokens and thus significantly more content than the 128,000 tokens mentioned in the video for ChatGPT. This makes it possible to analyze several log files at once.
Does my data leave the premises when I use Ollama?Ollama runs models locally, so the data normally does not get out. Nevertheless, verify this yourself in your environment, for example with a network monitor.
Which model is used with Ollama in the video?In the video, Llama 3.2 from Meta is used, which is available for free and as an open-source model.
What do enterprise solutions such as Azure OpenAI offer compared to the standard versions?They offer contractually guaranteed data protection, audit logs, compliance reports, SLAs, and integration into existing security infrastructures.
Do I need to consider anything before using an AI tool?Yes, first check your company's policies. Many companies specify which tools are permitted and which data may be uploaded.
