Recommended Free Tools
Generative AI makes phishing easier to scale and harder to dismiss on wording alone, and the defenses that hold up now work in several places at once: email filtering and detection, identity and configuration controls, and the human steps that decide what happens after a suspicious message is opened. Think of it as an arms race with uneven evidence. Attackers use generative AI to target better and send more. Defenders use detection and response technology alongside identity controls and processes. The published data supports that framing, but it does not support precise claims about who is winning.
Two recent publications show both sides. Microsoft’s Digital Defense Report 2025 compares click-through rates for AI-automated and standard phishing emails. An FBI public service announcement dated May 21, 2026 describes a phishing operation that uses AI-generated lures to capture Microsoft 365 OAuth tokens. Both are important, and both are narrower than the headlines that usually cite them.
As an Amazon Associate I earn from qualifying purchases.
What the published AI phishing numbers measure
Most figures in this area come from vendors, governments and complaint databases, and they measure different things. The table sets out what each figure covers and what it cannot tell you. Read it before using any number in a risk assessment or a board briefing.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute| Figure | Publisher and period | What it measures | What it does not show |
|---|---|---|---|
| 54% click-through for AI-automated phishing emails, compared with 12% for standard attempts | Microsoft, Digital Defense Report 2025 | Click-through rates Microsoft observed for the two categories of email in its own reporting | A universal benchmark; the share of all phishing that is AI-generated; whether any click led to a compromise. Sample size and observation window: not stated (Microsoft, 2025). |
| 4.5 times the click-through rate | Derived from Microsoft’s 54% and 12% figures | The arithmetic ratio of the two rates above | A separate measurement. It carries every limit of the rates it is calculated from. |
| 22,364 AI-related complaints and nearly $893 million in losses | FBI Internet Crime Complaint Center (IC3), 2025 data, reported April 2026 | U.S. complaints that IC3 classified as AI-related, and the losses reported with them | Email-only figures. The complaint data includes scams beyond email. |
| Nearly $21 billion in losses | FBI IC3, 2025 data, reported April 2026 | Losses from all cyber-enabled crime reported to IC3 in the U.S. | A phishing-only total, or a figure for any other country |
The 54% figure is the one most often quoted, and it is worth reading for what it implies rather than what it proves. It shows that lure quality matters to recipients. It does not show that most phishing now succeeds, and it does not predict how many people in your organization would click. The practical conclusion is to plan for clicks happening, not to assume that training or wording checks will prevent every one.
#1 Best Overall
What GAO says generative AI can and cannot do
The U.S. Government Accountability Office (GAO), in its Science & Tech Spotlight on the malicious use of generative AI, makes two points that frame everything else here. The first is about capability: “For example, paired generative AI systems could autonomously create and deliver phishing emails.” The second is about safeguards: “Even with safeguards in place, no current generative AI systems are immune to such misuse.” GAO also notes that attackers can manipulate safeguards through a variety of techniques, so defending against misuse requires continuous development and sustained resources.
That reasoning also explains why the evidence does not support a claim that most phishing is now written by AI. The published material describes what AI makes possible and documents specific cases. It does not establish a reliable share of all phishing that AI produced.
Can AI-written phishing get past email filters?
The reporting cited here does not measure how often AI-written messages get past filters, so no published figure answers that question directly. What changes is the signal that filters and people have relied on. Spelling and grammar errors were once common in phishing and still appear in many campaigns, but they can no longer be treated as a reliable test. A clean message may be malicious, and a message with errors may be legitimate.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
- Pass the Securing Email with Email Security Appliance 300-720 SESA with updated flashcards packed with detailed content aligned to the latest exam blueprint. Cover all core topics without the overload found in lengthy study guides. Get 300+ Securing Email with Email Security Appliance 300-720 SESA flashcards on 8-1/2″ x 11″ perforated card stock.
Microsoft frames the change as added scale and efficiency, with AI enabling more targeted phishing and better lures. The practical consequence is that defenses need to judge more than wording. They need to assess who sent the message and whether that sender is authenticated, what the message asks the reader to do, and what the account does afterward.
How defenders use AI-driven detection, and where it stops
The reporting describes the defensive side in general terms rather than as a survey of what companies deploy. Microsoft’s recommended responses and the FBI’s identity guidance point to detection and response that spans email, identity and collaboration tools, with the ability to correlate what happened across them.
Detection can be AI-based, and it is reasonable to use it to classify messages at volume and to surface sequences that a person would miss. But an AI detector alone cannot solve phishing. Detectors score content, while the damaging step is often a legitimate-looking request to approve a code, join a session or install a tool. No content score stops a user from approving that request.
Rank #3
- Pass the Securing Email with Email Security Appliance with updated flashcards packed with detailed content aligned to the latest exam blueprint. Cover all core topics without the overload found in lengthy study guides. Get 300+ Securing Email with Email Security Appliance flashcards on 8-1/2″ x 11″ perforated card stock.
The published evidence does not establish how accurately any particular AI-based detector performs against AI-written phishing. Claims that a product stops AI phishing should be treated as vendor assertions until independent, documented results are available.
How one email can start a longer attack
Microsoft’s report describes email as the opening step in a sequence rather than a single lure. The sequence it describes runs as follows:
- Flood the inbox. The attacker signs the target up for large volumes of newsletters or services, which Microsoft describes as email bombing.
- Hide the signals that matter. The flood can bury MFA prompts, password resets, fraud alerts or transaction notifications, so a genuine security warning gets missed.
- Make contact through a trusted-looking channel. A fake IT-support contact follows, and Microsoft reports that it can lead to installation of a remote access tool.
- Impersonate or deliver malware. Microsoft also describes impersonation and malware delivery as parts of these campaigns.
Microsoft does not give a frequency for this sequence in the reporting cited here, so treat it as a documented pattern rather than a measured rate. The practical lesson is that a single mailbox alert can look harmless. A surge of messages, a support call and an unexpected remote session are only obvious when they are read together.
Rank #4
- XGS 108 with 1 Year Xstream Protection - Next-generation firewall appliance with Xstream Protection subscription providing zero-day defense, cloud sandboxing, email filtering, intrusion prevention, and advanced reporting, managed through Sophos Central for unified policies and reporting.
- 6 x 2.5 GE copper ports and 1 SFP fiber port, supporting up to 12.5 Gbps firewall performance for growing business networks.
- Zero day protection with cloud sandboxing, email filtering, and advanced reporting for full enterprise coverage.
- TLS inspection and next generation intrusion prevention block hidden threats in encrypted traffic and stop sophisticated attacks.
- Includes Xstream Protection – Advanced security bundle with zero-day protection, cloud sandboxing, email filtering, and automated threat response, providing full coverage against the most sophisticated cyberattacks.
How a convincing email leads to Microsoft 365 account takeover
The FBI’s Kali365 public service announcement, dated May 21, 2026, describes a phishing-as-a-service operation that uses AI-generated lures and a device-code phishing flow to capture OAuth access and refresh tokens. In the flow the alert describes:
- The lure impersonates a trusted cloud or document service, so the target expects a routine authorization step.
- The attacker starts a device-code sign-in and asks the target to enter the code on a legitimate Microsoft verification page.
- The attacker captures the tokens issued by that authorization, including OAuth access and refresh tokens.
- The tokens provide access across Outlook, Teams and OneDrive. According to the FBI, that access can persist without another password or MFA challenge.
Because the page is genuine, checking only that the web address belongs to Microsoft will not reveal the problem. The danger lies in the authorization itself, which is why the response belongs in identity configuration as well as in user awareness.
Free tools Windows power users keep installed
One-click scans. No signup required.
The alert describes this operation specifically. It does not say that every phishing campaign uses device-code flow, so whether your tenant is exposed depends on whether that flow is used in your environment. The controls below address exactly that question.
Best Value
- XGS 88W with 1 Year Xstream Protection - Next-generation firewall appliance with Xstream Protection subscription providing zero-day defense, cloud sandboxing, email filtering, intrusion prevention, and advanced reporting, managed through Sophos Central for unified policies and reporting.
- Built in Wi Fi 6 with 4 x 2.5 GE copper ports, delivering up to 9.9 Gbps firewall performance for secure wired and wireless networks.
- Zero day protection with cloud sandboxing, email filtering, and advanced reporting for full enterprise coverage.
- TLS inspection and next generation intrusion prevention block hidden threats in encrypted traffic and stop sophisticated attacks.
- Includes Xstream Protection – Advanced security bundle with zero-day protection, cloud sandboxing, email filtering, and automated threat response, providing full coverage against the most sophisticated cyberattacks.
What organizations should do about AI-powered phishing
The controls below follow the recommendations in the reporting. They are a starting structure rather than a complete program, and they should be adapted to business needs.
Restrict device-code flow, after auditing dependencies
The FBI advises that restricting device-code flow to limit or block device authentication codes can help prevent or limit this style of attack. As described in the alert, the recommended steps are:
- Use a conditional access policy to block device-code flow where your environment allows it.
- Before enforcing the block, audit which legitimate apps, devices and processes depend on device-code sign-in, because blocking it can break them.
- Keep limited, documented exceptions where a legitimate dependency requires them.
- Consider blocking authentication transfer policies, which the FBI lists among related measures.
- Plan the change so that emergency-access accounts are not locked out by the new policy.
In Microsoft Entra Conditional Access, device-code flow is one of the authentication flows a policy can target. Menu labels change over time, so check them in your own tenant before editing a policy.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchFilter and monitor the sequence, not just the message
- Filter inbox floods. Microsoft recommends filtering the volume of mail that can bury security alerts, so genuine warnings stay visible.
- Correlate events across systems. A newsletter surge, a helpdesk contact and a new remote session should be investigated together rather than as separate tickets.
- Watch for account-compromise signals. Useful detections include suspicious OAuth authorizations, inbox-rule changes and impersonation of trusted senders.
- Keep detection as one layer. Monitoring supports the identity controls above. It does not replace them.
Limit collaboration and remote-access paths
- Control external Teams exposure. Review which outside accounts can contact staff, since Microsoft lists this among its suggested responses.
- Restrict remote monitoring and management (RMM) tools to the teams that need them, and alert on new installations.
Train people on the lures that are actually used
- Teach staff to recognize fake IT-support scams. Microsoft specifically recommends this training. A support contact that asks for a remote session, a code or a password should be verified through a known internal channel.
- Treat unexpected authorization prompts as suspect, including requests to enter a code on a Microsoft page.
- Make reporting quick. Suspicious messages should reach the security team while the sequence is still short.
How to evaluate email security products
The public reporting cited here does not include an independent comparative test that ranks named email security or identity products. Vendor marketing claims are not equivalent to controlled comparative results, so the sensible approach is to evaluate options against the same criteria rather than rely on a ranking.
Quick Recap
| Criterion | Questions to ask | Evidence to request |
|---|---|---|
| Platform compatibility and deployment | Which mail platforms, identity providers and collaboration tools are supported, and how is deployment done? | Documented supported configurations and deployment steps |
| Detection coverage | Does detection span email, identity, collaboration tools and post-delivery response? | Architecture description and examples of detections across each layer |
| Account compromise and impersonation | Can the product flag suspicious OAuth authorizations, inbox-rule changes and impersonation? | Documented detection list with descriptions of what each detection covers |
| Device-code and conditional access controls | Can administrators see and control device-code flow and related conditional access policies? | Administrator documentation for those controls |
| False positives and investigation context | How are false positives handled, how are user reports routed, and what context does an analyst receive? | Workflow documentation and the false-positive handling process |
| Data handling and retention | Where is mail processed, how long is it retained, and who can access it? | Data processing terms and retention settings |
| Evaluation methodology | What independent or controlled evaluation exists, and under what conditions, population and date? | The methodology, conditions and date of any published result |
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
