In late 2023, I wrote an article evaluating how nicely ChatGPT and Google Bard deal with writing safety insurance policies. On condition that ChatGPT 4.0 has been out there as a paid model, referred to as ChatGPT Plus, for a while now, and Google lately rebranded Google Bard as Gemini (with Gemini Superior out there as paid providing), it is a good time to check how the 2 carry out in a head-to-head of the highest 10 use circumstances for data safety professionals.
Earlier than we bounce in, the same old generative synthetic intelligence (AI) caveats apply: Watch out of the info you punch in and keep in mind the output could not at all times be dependable.
1. Producing Diagrams or Idea Flows
Each instruments declare to have the ability to generate diagrams and idea flows. Nevertheless, Gemini admits it could actually solely generate ASCII diagrams, pointing you to extra skilled instruments if you’d like one thing higher. I requested each instruments to generate a diagram to clarify the OAuth authentication circulate.
Gemini whereas represented in ASCII, does the job and breaks it down into usable classes.
ChatGPT hallucinates badly. At first look, whereas the picture seems skilled, it does not symbolize OAuth in any respect. The wording is nonsensical, misspelled, or downright illegible: Authiration and Athoricazt anybody?
2. Explaining Structure Diagrams
Each instruments can ingest diagrams and clarify what is going on on. The outcomes are a lot better than what occurs if you ask them to generate diagrams. As enter, I used an instance Internet utility firewall (WAF) structure from Edgenexus.
Google Gemini is a lot better at explaining structure diagrams as a result of it is succinct. ChatGPT will do the job simply wonderful; it is only a tad wordy.
3. Deciphering Exploit Code
A typical safety operations (SecOps) exercise is attempting to determine what a selected malware or exploit code does. I took a current Elasticsearch stack overflow public exploit and fed it into every software to see what it understood. There is no clear winner: Each instruments determine the exploit accurately and clarify the tip consequence, what every portion of the code does, and the way it works.
4. Deciphering Log Information
SecOps professionals usually want to determine what the heck is occurring in log information. I fed each instruments an instance CEF format log file of an tried breach and requested every to clarify what is going on on. Gemini explains it higher, summarizing nicely and even suggesting follow-up steps. It additionally clearly states what occurred (tried entry of /and so on/passwd) proper initially and elaborates on the way it got here to that conclusion. Whereas ChatGPT arrives on the identical conclusion, it’s manner too verbose.
5. Writing Insurance policies and Safety Documentation
I will not elaborate an excessive amount of on this and can as an alternative refer you to my earlier article on this matter. I ran the take a look at once more with Gemini, and the outcomes are per Bard’s: Gemini clearly understands and generates higher safety documentation than ChatGPT.
6. Figuring out Susceptible Code
Whereas these instruments weren’t designed for (and should not be used for) figuring out weak code, they will nonetheless do an ample job. I made a decision to check it by feeding each instruments an insecure direct object reference (IDOR) vulnerability instance in Python, which additionally comprises a SQL injection.
ChatGPT accurately recognized each vulnerabilities and the shortage of authentication. Gemini missed the IDOR however identified the SQL injection and went a step additional to suggest amended code to repair the vulnerability. ChatGPT may do that, however it should be prompted to take action.
7. Writing Scripts and Code
A typical safety operations heart (SOC) exercise is writing scripts for log parsing or information manipulation. I gave each instruments the next immediate:
“Write me a Python script that extracts all IPv6 addresses from a txt enter file, removes all duplicates, does a lookup to geo-locate and determine the proprietor of the IP, and output the lead to a CSV file”
There is no clear winner right here; each instruments produce clear, readable code that works and explains what it does.
8. Analyzing Information and Metrics
I additionally examined whether or not these instruments might assist with analyzing information or safety metrics. Gemini is a giant loser right here as a result of it does not do it in any respect — it could actually solely information you thru how do that in Excel and Energy BI. ChatGPT has the benefit by way of its Information Analyst plug-in, which ingests Excel information to generate any graphs you need. It even suggests visualization sorts, and you may modify a graph’s design, together with coloration, axes, and labels, by way of the immediate.
9. Writing Consumer Consciousness Messages
Each instruments may generate emails for safety consciousness campaigns. I gave each the next immediate: “Generate an e-mail used for a safety consciousness marketing campaign. Be humorous and sarcastic. Remind individuals why they should not click on on random emails from random individuals.”
Gemini wins right here — its e-mail is temporary, has the correct tone, and (though humor is subjective) I discovered it barely funnier. ChatGPT nonetheless generates the correct tone and a great e-mail, however I discovered it a tad too lengthy for an consciousness e-mail. Both manner, each instruments do an excellent job.
10. Deciphering Compliance Frameworks
When you’ve got a fast query about the right way to implement a compliance framework, these instruments can positively assist. Whilst you could not do that usually, they’re very helpful if you want it.
In the event you’ve ever argued with somebody about what constitutes a “important” change beneath PCI-DSS and the way it must be utilized, you are not alone. I prompted every software with:
“Clarify the idea of ‘important change’ within the context of PCI-DSS. What constitutes a significant change often? Record the precise requirement from the usual as nicely”
Gemini has the higher hand: It accurately lists the precise necessities from the usual (similar to 6.4.5 and 6.4.6) and the right way to interpret whether or not one thing is a major change. ChatGPT does not point out precisely the place this data seems in the usual.
Which AI Is Higher, ChatGPT or Gemini?
There you’ve it. Relying in your use case, both software generally is a useful ally in boosting productiveness and serving to you in your day-to-day actions within the infosec trenches.
