Phishing detection got a lot of attention at RSA Conference 2025. Multiple vendors showed off new tools. Researchers presented new techniques. And the message was clear: the old ways of catching phishing emails are not keeping up.
That matters to you because phishing is still the number one way attackers get into your environment. It has been for years. And it is not slowing down.
The Bottom Line
If you just want the highlights, here they are:
- AI-driven phishing attacks are outpacing traditional email filters
- New detection systems are using behavioral analysis and large language models to fight back
- Real-time link analysis and brand impersonation detection are becoming standard features
- Most of these advances will show up in paid platforms — budget accordingly
- Your people are still your last line of defense, and training has to evolve too
What Is Actually Changing
Traditional phishing detection relies heavily on signatures and reputation lists. Think of it like a bouncer at a bar with a list of known troublemakers. If your name is on the list, you do not get in. If you are new, you walk right through.
That model is broken. Attackers now use generative AI to craft emails that are grammatically perfect, contextually relevant, and free of the obvious red flags we trained people to spot. No more Nigerian prince emails with misspelled words. These messages look like they came from your CEO’s actual inbox.
The new detection systems presented at the conference take a different approach. Instead of just checking a list, they analyze behavior. They look at how an email was constructed, where it actually originated, whether the sending patterns match historical norms, and whether the language in the message is consistent with how the supposed sender typically writes.
Some systems are now using large language models — the same technology behind ChatGPT — to evaluate whether an email’s content is likely machine-generated. That is a meaningful shift. You are essentially using AI to catch AI.
Real-Time Link and Brand Analysis
Another area getting major upgrades is link analysis. Old-school systems check a URL against a blocklist. New systems actually follow the link in a sandboxed environment, render the page, and evaluate it in real time.
They look at the visual layout of the destination page. They compare it against known brand assets — logos, color schemes, login page structures. If someone cloned your bank’s login page and hosted it on a throwaway domain registered six hours ago, these systems catch it.
Brand impersonation detection is also getting smarter. Instead of relying solely on domain name checks, newer tools analyze the full context of the message — the tone, the branding elements embedded in the email body, even the favicon on the linked site.
What You Should Do
- Evaluate your current email security stack. If your phishing protection is still primarily signature-based, you are behind. Ask your vendor specifically what behavioral and AI-driven detection capabilities are included in your current license.
- Ask about LLM-based detection. This is the new frontier. Find out if your email security provider has a roadmap for detecting AI-generated phishing content. If they look at you blankly, that tells you something.
- Enable real-time link scanning. Many platforms offer this but it is not always turned on by default. Make sure URLs are being detonated and analyzed at click time, not just at delivery.
- Update your security awareness training. Your training program should reflect the new reality. Show your team examples of AI-generated phishing emails. The old “look for typos” advice is dangerously outdated.
- Budget for upgrades. My guess is many of the most impressive capabilities shown at the conference will be foundational to paid product tiers. The free or base-level protections are not going to cut it. Plan your spend accordingly.
A Word of Caution
Conference demos are designed to impress. Vendors show you the best-case scenario under controlled conditions. Real-world performance is always messier.
Do not rip and replace your email security based on a flashy demo. Run a proof of concept. Test against your actual mail flow. Measure false positive rates, because a system that quarantines legitimate business email is a productivity killer that your users will learn to ignore.
The Bottom Line
Phishing detection is getting meaningfully better. That is the good news. The bad news is that it has to, because the attacks are getting better too. This is an arms race, and standing still means falling behind.
Review your defenses now. Not next quarter. Because the phishing email that gets through tomorrow will not have a single typo in it.











