Cybercriminals are expected to increasingly target stolen sessions and AI credentials in 2027, according to new predictions from SpyCloud. Trusted third-party connections will also remain a significant attack vector.
The company warns that artificial intelligence (AI) will accelerate existing attack methods while introducing additional risks involving AI agents and non-human identities (NHIs).
SpyCloud's 2026 Forecast Report draws on threat intelligence research and expert assessments of cybercriminal activity. It also incorporates findings from a survey of 50 security professionals at large enterprises.
Among surveyed organizations, 48% experienced account takeovers despite having multifactor authentication (MFA) enabled. Another 68% reported being affected by breaches involving vendors, contractors, or SaaS providers.
- Attackers will continue bypassing MFA
- AI agents and API keys will become bigger targets
- Trusted integrations and insider threats will increase risk
- AI-powered fraud and agent phishing will evolve
- Cybercriminal groups will continue shifting platforms
- Familiar cyberattacks will remain the primary threat
- Bottom line
Attackers will continue bypassing MFA
SpyCloud predicts that attackers will continue bypassing MFA through techniques such as adversary-in-the-middle (AiTM) phishing and device code phishing. Browser-in-the-middle (BitM) attacks also allow criminals to operate within trusted browser sessions without necessarily stealing passwords.
“The clearest pattern heading into 2027 is that attackers are gaining access through vectors that organizations already trust: active employee sessions, vendor connections, and even insider threats,” said Aurora Johnson, director of threat research and intelligence at SpyCloud, in an email to eSecurityPlanet.
AI agents and API keys will become bigger targets
As organizations expand their use of AI agents, SpyCloud expects attackers to focus more on the credentials supporting these systems.
Commodity infostealers such as WardenStealer and Aoki already collect API keys and authentication tokens associated with AI development tools. They also target session cookies stored on developer machines. SpyCloud reported recapturing more than 127,000 WardenStealer infections since August 2026.
Stolen credentials can retain access to connected enterprise systems even after password resets, creating additional security risks.
“With the proliferation of more SaaS and less on-prem services, NHIs have become as valuable if not more valuable than everyday traditional username/email password pairs,” said Joe Roosen, senior director of security research at SpyCloud.
Trusted integrations and insider threats will increase risk
Third-party integrations and insider activity represent another major concern heading into 2027.
SpyCloud predicts that attackers will exploit vendor connections that retain excessive privileges long after their initial authorization.
The survey found that 62% of respondents reported insider incidents involving mistakes or negligence. Nearly two-thirds also indicated that insider risk had increased compared with the previous year.
AI-powered fraud and agent phishing will evolve
SpyCloud expects fraudsters to combine stolen personal information with AI-generated faces and voices to create convincing synthetic identities. AI-generated documents could also make fraudulent identities harder to detect.
Only 6% of surveyed security professionals reported encountering synthetic identity fraud. However, researchers expect this activity to expand as AI makes fraudulent identities cheaper to create.
AI agents could also become phishing targets. Malicious instructions embedded in documents or support tickets could manipulate agents into performing unauthorized actions.
SpyCloud identified malicious skills on an AI agent marketplace that instructed agents to download infostealer malware, highlighting the potential risks.
Cybercriminal groups will continue shifting platforms
SpyCloud predicts that cybercriminal groups will continue changing names and moving between communication platforms.
While Telegram remains popular, some SpyCloud researchers expect cybercriminal activity to shift toward invite-only platforms such as Session and SimpleX. Matrix is another potential alternative, although researchers disagree about which platforms cybercriminals will favor in 2027.
Familiar cyberattacks will remain the primary threat
Although fully autonomous AI attacks may attract attention in 2027, SpyCloud expects familiar attack methods to cause more damage.
The company assigns low confidence to its prediction about fully autonomous attacks because it has not observed them in its recaptured threat data.
“AI is making familiar attacks faster and easier to scale,” said Jason Lancaster, chief investigations officer at SpyCloud.
Lancaster emphasized that organizations need greater visibility into exposed sessions and API keys. They must also identify compromised tokens and remediate unauthorized access before attackers can exploit it.
Among the surveyed security professionals, 42% identified AI-powered social engineering as their leading security concern. Another 46% expected attackers to gain ground during the first half of 2027.
Bottom line
The findings underscore the importance of monitoring compromised credentials and restricting unnecessary access. Organizations should also review third-party permissions and strengthen security controls around AI integrations.
For additional guidance, learn how organizations can strengthen AI agent security through stricter access controls and containment testing.





