About
Subscribe

ANALYSIS: AI app boom exposes cracks in the code

Nicola Mawson
By Nicola Mawson, Contributing journalist
Johannesburg, 02 Nov 2026
Out of more than 9 000 AI-generated open-source applications, nine out of 10 contained at least one security vulnerability. (Image created with GenAI)
Out of more than 9 000 AI-generated open-source applications, nine out of 10 contained at least one security vulnerability. (Image created with GenAI)

Artificial intelligence (AI) app use is skyrocketing, with more time than ever spent on GenAI apps – projected to reach 36 billion hours in the first half of the year.

The release of AI-developed apps has doubled year-on-year, despite nine in 10 of AI-generated apps in one survey containing at least one .

Some 90% of professional developers were using AI coding agents at work at least weekly in one form or another (local agents or remote cloud agents), with 68% using them daily between May and July, JetBrains Research finds in an August report on its Developer Ecosystem Survey 2026.

provider Sensor Tower states in its State of AI 2026 report that consumer spending on AI is hitting record highs, driven by a shift toward premium, utility-driven subscriptions. Global in-app purchase revenue from AI apps is expected to surpass $4 billion in the first half, an increase of 36% over the second half of 2025, it finds.

Nearly 1.3 million new iOS and Google Play apps were released in the first half of 2026 alone, double the amount seen in either the first half of 2025 or that of 2024, The Atlantic reports, quoting Sensor Tower.

The popular kid

Oliver Yeh, co-founder and CEO of Sensor Tower, notes: “US app downloads driven by ‘AI’ keyword searches have skyrocketed over the past two years, and apps featuring ‘AI’ in their descriptions are on track to reach 10 billion global downloads” in the first half of 2026 alone.

Sensor Tower predicts that global time spent on GenAI apps will hit 36 billion hours in the first six months of the year – more than double that of the first half of 2025. “These numbers make it clear that AI is no longer just reshaping how technology is built; it’s reshaping how consumers discover, engage and spend,” says Yeh.

“Sensor Tower data shows the rapid adoption of AI by marketers, advertisers and app developers across sectors in an effort to better serve consumers and drive engagement.”

AI tools may increase developers’ coding activity by up to 180%, but the gains translated into a 50% increase in projects and just 30% more software releases compared with developers not using AI, the MIT Sloan School of Management notes.

The race is on

Sensor Tower also states that ChatGPT still leads what it calls the AI assistant market. It is, it says, “becoming the fastest mobile app ever to reach one billion monthly active users in May 2026,” a milestone it achieved in just three years.

“More broadly, the market remains highly concentrated, with ChatGPT, DeepSeek and Google Gemini accounting for nearly 90% of total time spent across AI assistant apps” in the first quarter of the year, says Sensor Tower.

ChatGPT broke records when it reached one billion monthly active users in May 2026. (Source: Sensor Tower, redrawn by GenAI)
ChatGPT broke records when it reached one billion monthly active users in May 2026. (Source: Sensor Tower, redrawn by GenAI)

Sensor Tower notes competition is intensifying, with ChatGPT use being challenged by the likes of Google Gemini and Claude. “Claude has experienced explosive growth, led by a strong web presence, and its True Audience share in the US has more than tripled,” it says.

The use of Claude Code is growing at an unprecedented rate and has become the most widely adopted AI coding tool at work, says JetBrains. “It is used twice as often as GitHub Copilot, a former long-standing leader of the market that brought AI-assisted coding into the spotlight in 2023,” it says.

But is it safe?

Yet, a paper entitled Understanding the (In)security of Vibe-Coded Applications, finds that out of more than 9 000 AI-generated open-source applications, nine out of 10 contained at least one security vulnerability. “Insecurity is the norm rather than the exception,” they write.

One of the outcomes of a study on whether vibe-coding is safe, based on authors benchmarking for the Vulnerability of Agent-Generated Code in Real-World Tasks report, notes that those tools frequently produce code with security vulnerabilities.

While 57% of the solutions generated by one agent using Claude Sonnet 4 worked as intended, only 11.8% were secure, the paper says, adding that including security warnings to task instructions did little to reduce the risks. The research was based on an evaluation of 12 AI coding agents against 186 software development tasks.

Share