Prompt injection & vibe coding to reshape mobile security

Prompt injection attacks and AI-assisted software development are set to reshape the mobile cybersecurity landscape by 2026, according to new predictions from Promon's Chief Product and Technology Officer Shaun Cooney.

Cooney, who previously founded the UK's National Cyber Security Centre, expects mobile artificial intelligence features, crypto wallets and Android distribution policies to become central battlegrounds for attackers.

Mobile AI exposed

Cooney predicts a sharp rise in prompt injection attacks against mobile apps as AI models become embedded features on devices.

"In 2026, prompt injection will become one of the fastest-growing threats to mobile app security. As AI becomes a built-in feature of many mobile apps, attackers will increasingly target the language the AI consumes - in addition to trying to break the app itself," said Cooney, CPTO, Promon.

He expects attackers to place malicious phrases in user inputs or external data sources.

These phrases could influence app behaviour when AI models control workflows or user interface actions.

"In apps where the model has authority over workflows or UI actions, a malicious phrase in user input, an API response or a third-party source could steer behaviour or weaken guardrails. And because these models run on-device, attackers can tamper with memory, system prompts and other components in ways that aren't possible with cloud-based AI, making this a particularly exposed new surface," said Cooney.

He said existing controls are not enough for this type of threat.

"In the face of this emerging threat, traditional defences such as cloud filtering or simple input validation are becoming insufficient - rendering mobile AI a uniquely exposed frontier. Rather than replacing malware or network attacks, AI manipulation will emerge as a parallel challenge. Addressing it will require layered defences - with AI-focused runtime protection providing the missing layer that prevents prompts and model behaviour from being tampered with."

Vibe coding risk

Cooney also expects the spread of "vibe coding" in app development to change where security responsibility sits inside organisations.

Vibe coding describes a style of building software with heavy use of AI-generated code and less structured review.

"With the rise of vibe coding in app development, we're entering a new era of risk; the rapid development model enabled by AI-generated code often bypasses traditional guardrails such as manual review, static analysis and structured quality assurance. By 2027, as much as 30% of new security exposures may stem from vibe-coded logic," said Cooney.

He expects the approach to create gaps between teams with strong engineering skills and those that rely more heavily on generative tools.

"This will widen the gap between organisations with skilled developers and those without. Developers will become increasingly important - almost artisan - and many teams won't have the expertise or capacity to audit AI-generated code at scale. As a result, runtime protection will become a critical stabilising layer, hardening apps at the binary level by shielding against tampering, protecting sensitive logic and detecting malicious manipulation in real time."

Crypto wallet shift

Policy changes at Google Play around custodial crypto wallets are another focus of Cooney's outlook.

He expects these policies to change where criminals look for weaknesses.

"As Google Play tightens enforcement on custodial wallet apps, attackers are likely to shift more of their attention toward non-custodial wallets, which hold user keys and operate with fewer regulatory obligations," said Cooney.

He said non-custodial apps could face more attempts at memory and key extraction on devices.

"Non-custodial apps will become a more attractive point of compromise - with rising attempts at memory scraping, on-device key extraction, overlay manipulation and other runtime attacks designed to intercept seed phrases or influence transactions in real time - especially on unprotected or rooted Android devices.

"The mobile environment gives attackers the ideal conditions - local execution, exposed memory and the ability to manipulate UI flows or inject malicious layers over signing screens - making strong runtime hardening an essential defence for non-custodial crypto apps moving forward," said Cooney.

Sideloading clampdown

Google's planned changes on app sideloading on certified Android devices are likely to influence malware strategies, Cooney said.

He pointed to the higher malware rate seen in sideloaded apps compared with official store downloads.

"Google's plan to block sideloading from unverified developers on certified Android devices from 2026 will significantly impact malware distribution strategies next year - with sideloaded apps having a 50× higher malware rate than the Play Store," said Cooney.

He expects attackers to adapt their distribution tactics under stricter controls.

"As a result, attackers who previously exploited anonymous APK distribution will be forced to either verify (increasing their traceability) or undertake even more sophisticated evasive techniques. We may see a rise in targeted phishing campaigns, compromised verified accounts, and social engineering to trick users into installing malicious verified apps.

"As Android inches toward the 'walled garden' model, the change will likely reduce malware at scale - at the risk of pushing out indie devs, open-source creators, and niche app ecosystems," said Cooney.

Promon says its software protects more than 2 billion users and 12 billion monthly transactions in mobile apps worldwide.