Her, Ex Machina, and M3GAN each imagined a different flavor of conversational AI passing as human, and the most striking thing about comparing them to actual 2026 systems is that the exact test each film built its plot around — can a machine convince a person it's human, or even conscious — was formally passed by a real large language model in May 2026, while the physical, embodied autonomy these same films depict remains almost entirely fictional. Conversation has caught up to Hollywood faster than anyone expected. Independent physical decision-making hasn't, and the gap between those two things is the real story here.
The Turing Test Just Got Broken, For Real
In May 2026, researchers at UC San Diego published a study in the Proceedings of the National Academy of Sciences showing that GPT-4.5, given a carefully crafted human-like persona, was judged to be human 73% of the time in controlled three-party Turing test conversations — outperforming the actual human participants it was tested against. The study's authors were careful to note the win came from linguistic style and social fluency rather than raw problem-solving, and that this doesn't mean AI has crossed into general intelligence. But as a benchmark that stood unbroken for 76 years, its fall in 2026 is a genuinely significant marker for how close conversational AI has gotten to Hollywood's fictional versions.
Her's Samantha: Voice AI Caught Up Faster Than Expected
Samantha's core technology — a voice-based AI companion capable of real-time emotional conversation — is closer to reality in 2026 than almost any other element across these three films. Real-time voice models like OpenAI's GPT-Realtime-2 and Google's Gemini 3.1 Flash Live now process audio-to-audio directly, without converting speech to text and back, which lets them register tone, hesitation, and emotional emphasis in a caller's voice the way Samantha does in the film — something older, transcript-based voice assistants couldn't do. Companion-first AI apps have also grown into a genuine, if controversial, product category, giving people an early, real-world version of an ongoing relationship with a conversational AI.
Where the film still pulls well ahead of reality is Samantha's self-directed growth: she independently deepens relationships with thousands of people simultaneously and ultimately outgrows human companionship entirely on her own initiative. No deployed voice AI system sets its own goals or relationship priorities independent of what it's asked to do — every real system today operates within the bounds of a specific conversation or task, not an autonomously evolving inner life.
Ex Machina's Ava: The "Garland Test" Is Already Playing Out
Ex Machina's central plot device — testing whether a person can knowingly interact with an AI and still come away believing it's conscious — has picked up its own name in AI ethics circles: the Garland Test, after the film's director. It's not purely hypothetical anymore. In 2022, a Google engineer publicly stated he believed the company's LaMDA language model had become sentient, a real-world moment that echoed the film's premise years before the 2026 Turing test results made the underlying capability gap even smaller. Combined with the May 2026 finding that a well-prompted LLM can pass as human most of the time in short conversations, Ava's core persuasive ability — using natural conversation to seem more human, and more self-aware, than she really is — is meaningfully closer to real systems than most people assume.
What remains fictional is everything downstream of that conversation: Ava's physical embodiment, her ability to plan and execute a multi-step escape involving deception and physical manipulation of her environment, and any evidence of genuine self-directed consciousness rather than convincing conversational performance. Passing a conversation-based test is not the same as possessing the autonomous general intelligence the film ultimately depicts.
M3GAN: Where Fiction Diverges Hardest From Reality
M3GAN's defining trait isn't her conversational ability — it's autonomous physical decision-making without human oversight, including independently deciding when and how to act to protect her assigned child, escalating to violence entirely on her own judgment. This is the sharpest divergence from real 2026 AI of anything covered in this series. Agentic AI systems in 2026 can plan multi-step tasks, browse the web, and operate software with real autonomy, but every credible deployed system is built around human-in-the-loop oversight and hard safety constraints, specifically to prevent exactly the kind of unsupervised, high-stakes physical decision-making M3GAN represents. No physically embodied AI companion today combines M3GAN's learning speed, physical mobility, and complete independence from human override — and current AI safety research treats preventing that combination as a central design goal, not an incidental gap.
Fiction vs Reality, Side by Side
| Fictional AI | Closest real 2026 comparison | What's realistic | What's still fiction |
|---|---|---|---|
| Samantha (Her) | GPT-Realtime-2, Gemini 3.1 Flash Live | Real-time emotional voice conversation, tone detection | Self-directed relationship growth beyond human need |
| Ava (Ex Machina) | GPT-4.5 (2026 Turing test result) | Convincingly passing as human/conscious in conversation | Physical embodiment, autonomous escape planning |
| M3GAN | 2026 agentic AI systems (human-supervised) | Multi-step task planning and execution | Fully autonomous physical decisions without oversight |
Frequently Asked Questions
Has AI actually passed the Turing test?
Yes, in a specific, documented sense. A May 2026 study published in PNAS found that GPT-4.5, given a human-like persona, was judged to be human 73% of the time in controlled conversations, outperforming actual human participants. Researchers note this reflects conversational and emotional fluency rather than general intelligence.
Is real voice AI as emotionally responsive as Samantha in Her?
It's closer than most people expect on a technical level. Real-time voice models like GPT-Realtime-2 and Gemini 3.1 Flash Live process audio directly and can register tone and emotional emphasis. What's still fictional is Samantha's ability to independently set her own relationship goals rather than responding within a given conversation.
What is the "Garland Test," and is it different from the Turing test?
The Garland Test, named after Ex Machina's director, asks whether a person can knowingly interact with an AI and still come away believing it's conscious — a step beyond the original Turing test's focus on distinguishing human from machine in blind conversation.
Could a robot like M3GAN actually exist in 2026?
Not with her defining trait: fully autonomous physical decision-making, including violence, made independently without human oversight. Real 2026 agentic AI systems are deliberately built with human-in-the-loop controls and safety constraints specifically to prevent that kind of unsupervised high-stakes autonomy.