Google is adding some new security features to Gmail, including improved phishing detection that relies on a dedicated machine learning model.
Anywhere from 50 to 70 percent of messages received in Gmail are spam, but machine learning techniques help Google bock spam and phishing messages with more than 99.9 percent accuracy, the company says.
Its latest machine learning model improves the process by delaying select messages (less than .05 percent of messages on average) for further phishing analysis.
Google says it's new detection models will also help generate new URL click-time warnings, giving users a warning prompt when they've clicked on a suspicious link. As new patterns are found, the models will adapt and get better with time.
Additionally, Google is adding unintended external reply warnings to Gmail to prevent users from accidentally sending protected data to someone outside of their company. Gmail shouldn't overdo it with the warnings, Google says, since its contextual intelligence will alert Gmail if the recipient is an existing contact or someone the sender interacts with regularly.
Google is also adding new built-in defenses against ransomware and polymorphic malware, blocking millions of additional emails. The threats are identified by combining signals from spam, malware and ransomware with attachment heuristics (emails that could be threats based on signals) and sender signatures (already marked malware).