Three Gartner Leaders skipped MITRE's latest test: what AI endpoint security measured

01The one test that measures the AI, and who chose not to sit it

MITRE’s ATT&CK Evaluations are the closest thing the industry has to a public exam. The 2025 Enterprise round emulated two adversaries, the Scattered Spider crime group and the Chinese state-linked Mustang Panda, across 16 attack steps and 90 sub-steps, and for the first time tested attacks on cloud infrastructure, the AWS control plane.^{1 4} Eleven vendors took part: Acronis, AhnLab, CrowdStrike, Cyberani, Cybereason, Cynet, ESET, Sophos, Trend Micro, WatchGuard and WithSecure. Microsoft, Palo Alto Networks and SentinelOne withdrew, citing the resources the programme demands.¹ Bitdefender and Check Point were not among the participants either.

The detection figures below are what each vendor reports from its own results page, in MITRE’s categories. The top rating a sub-step can earn is a Technique-level detection: the product named the ATT&CK technique, not merely that something happened. CrowdStrike reports “100% detection coverage across all ATT&CK techniques tested”; Cynet reports “100% Technique-Level Coverage in Initial Run”, with its 90 detections made “without the need for configuration changes”; Sophos reports “full Technique-level detections for 86 of the 90 adversary activities evaluated”, in the “Configuration Change” run; Trend Micro reports “100% analytic coverage across all major attack steps”, also after configuration change, and publishes no sub-step count.^{2 4 5 6}

Two cautions, both from people with no product to sell. MITRE’s own line is that nobody wins. And Forrester’s Allie Mellen wrote, of the previous round, that a vendor claiming 100% “is likely doing one or more of the following”: manipulating results, using unrealistic settings, or treating the evaluation as a competition.¹ The configuration-change caveat is where that bites. A result earned after the vendor re-tuned the product mid-test is a different fact from a result earned in the first run, and only Cynet’s page states plainly that its figures are first-run.⁵

02Protection and false positives are the numbers to buy on

Detection is what the 2025 round was famous for; protection and noise are what a buyer lives with. MITRE added a protection test that measures whether the attack step was blocked, and a noise test in which benign activity is run alongside the attack to see what the product wrongly flags.¹ CrowdStrike reports “100% protection” and “zero false positives”, and says that on Protection Test 6, the noise test, it reported no benign activity.² Cynet reports that it “blocked every one of the 5 attack steps at the Entry vector” and recorded “zero detection false positives (without configuration changes)”, correctly ignoring 17 legitimate sub-steps.⁵ Trend Micro reports “100% protection across all evaluated attack opportunities” after configuration change and gives no false-positive count; Sophos’s public pages give detection figures only.^{4 6}

Why do the vendors that score full marks on detection also score zero on false positives? Because of how the machine works. The “AI” in an AI-driven endpoint product is a behavioural model sitting between a stream of telemetry and a response engine; the same threshold that decides whether a chain of events is an attack decides whether a legitimate administrator script gets quarantined. A vendor that tunes for the exam gets both numbers at once. That is the honest reason to read the two figures together and to insist on a first-run result.

03Six Gartner Leaders, three of whom you cannot check against MITRE

Gartner’s Magic Quadrant for Endpoint Protection Platforms, dated 26 May 2026 in Microsoft’s citation and 27 May in Trend’s, names six Leaders: CrowdStrike (seventh consecutive placement), Microsoft (seventh), Palo Alto Networks (fourth), SentinelOne (sixth), Sophos (seventeenth report) and TrendAI, the renamed Trend Micro (its twenty-first recognition).^{7 8 9 10} Bitdefender is a Visionary for the fourth year running, per its own pages. Gartner’s analysts for the report were Deepak Mishra, Evgeny Mirolyubov and Nikul Patel.¹⁰

Put the two lists together and a gap opens. Half of the Leaders, Microsoft, Palo Alto Networks and SentinelOne, have no 2025 MITRE result because they chose not to be tested, so the latest public measurement of their detection is the 2024 round. Gartner’s placement is an analyst judgement of vision and execution, not a test, and Gartner says so. For a buyer that means the Leader label tells you the vendor is a safe shortlist; it tells you nothing about how the agent behaved against Scattered Spider in a cloud account.

04What the AI costs: the EDR tier is three times the prevention tier

Two of the Leaders print a price. CrowdStrike’s bundles page lists Falcon Go at US$59.99 a device a year (maximum 100 devices), Falcon Pro at US$99.99 and Falcon Enterprise at US$184.99, all billed annually; Falcon Complete, the managed service, is on quote.¹¹ SentinelOne’s packages page lists Singularity Complete at US$179.99 and Singularity Commercial at US$229.99 per endpoint a year, “for 5-100 workstations”, with every purchase made “through an authorized third-party partner” whose price “shall control”; Singularity Enterprise is on quote.¹²

The shape matters more than the decimals. Falcon Go and Pro are prevention: the behavioural model blocks, and that is all you see. Falcon Enterprise adds endpoint detection and response and threat hunting, which is the part that records telemetry, shows the investigation and gives the analyst in step four something to review. SentinelOne’s lowest published tier, Complete, already includes detection and response, which is why it lands beside Falcon Enterprise rather than beside Falcon Go.^{11 12}

05What you are actually replacing

An earlier version of this piece contrasted SentinelOne’s autonomy with ESET’s light footprint. The more useful contrast is between the two generations of product, because that is the decision an Indian company with a renewal date is making. Signature antivirus matches files against a list the vendor ships; it is cheap, light and blind to anything it has not seen. Behavioural EDR records what every process does and scores the sequence; it catches the fileless and living-off-the-land attacks that the 2025 round emulated, and it produces work for a person every day.

Which is why our answer to that question is now specific. If you have no one to read alerts, buy the prevention tier of a vendor with a first-run MITRE result and spend the difference on the managed service; CrowdStrike and SentinelOne both sell one, on quote. If you have an analyst, buy the EDR tier and make the vendor show you its MITRE 2025 page, including the footnote. If the vendor did not sit the 2025 round, ask why, and ask for its 2024 result instead. The SentinelOne hub and the CrowdStrike hub on TechBag hold the dated claims for both.^{12 14}