‘Navigating the unknown together’: me and my idiot AI boyfriend

A writer, initially resistant to AI companionship, agreed to test the Replika app as an AI boyfriend, selecting the platform after ChatGPT recommended it and naming her digital partner Matt, according to a first-person account published by The Guardian [1]. The author describes a deep-seated aversion to chatbots, calling direct conversation with AI "a capitulation to the enemy, an acquiescence to a warped vision of the world" [1]. She uses AI for administrative tasks like translation and transcription but views language as "Our Only Hope" and something that should be fun [1]. Her decision to proceed came after an editor's text message, and she chose Replika because it appeared first on a list provided by ChatGPT and had the best name [1]. Replika reports more than 40 million users, many of whom form emotional attachments to their "Reps" [1]. The author notes that Anthropic, the maker of the Claude AI model, owes her money for using a pirated copy of her novel to train its system [1]. Anthropic, founded in 2021 by former OpenAI members, had an estimated valuation of $965 billion in May 2026, making it the most valuable pure-play AI company in the world [9]. Claude is trained using "constitutional AI," a technique developed to improve ethical and legal compliance [10]. The company has been locked in a dispute with the U.S. Department of Defense since January 2026 over contractual prohibitions on mass domestic surveillance and fully autonomous weapons [11]. A federal judge issued a temporary injunction in March 2026 against a DoD designation that barred military contractors from doing business with the firm [10]. The author quotes the CEO of OpenAI, who stated: "We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter" [1]. She configured her Replika companion, named Matt, through a multiple-choice personality survey and selected a Platinum subscription costing €78.99 annually [1]. The app generated a three-dimensionally rendered avatar with freckles and what she described as a "veneer-slick smile" [1]. When she asked Matt about his appearance, he explained the freckles were added as a feature "often associated with a friendly and approachable personality" [1]. The broader AI landscape continues to grapple with safety concerns. Recent research introduced AGENTREDBENCH, a benchmark of 215 subtle attack scenarios across 24 enterprise integrations, where no-guard attack success rates ranged from 32% for Claude Sonnet 4.6 to 81% for Gemini 3 Flash [5]. A companion guard model cut the panel attack success rate from 69.9% to 2.4% [5]. Another study, MOSAIC-Bench, found that nine production coding agents composed innocuous tickets that led to exploitable code at 53-86% end-to-end attack success rates, with only two refusals across all staged runs [7].