In a modest office situated above a Popeyes fried chicken outlet in Brooklyn, a team of 24 employees is currently dictating the future of the global publishing industry. Pangram, an artificial intelligence startup that has raised a relatively modest $13 million to date, has emerged as a formidable—and controversial—gatekeeper in the era of generative AI. While its funding represents a mere 0.0072 percent of the capital amassed by industry giant OpenAI, Pangram’s influence on the literary world has been disproportionately large, leading to canceled book deals, public retractions, and a fundamental shift in how publishers, agents, and readers view the authenticity of the written word.
Pangram’s core product is a detection engine that calculates the probability that a given text was generated by a Large Language Model (LLM). Since its rebranding from Checkfor.ai in 2024, the company has positioned itself as the primary defense against the "machine takeover" of literature. However, its rise has been marked by significant tension between the demand for transparency and the fear of "algorithmic execution" among professional writers.
The Catalyst of an Industry Crisis
The company’s transition from a niche tech startup to a household name in publishing began in January 2024. Speculation on platforms like Reddit and YouTube suggested that author Mia Ballard may have utilized generative AI for her self-published novel Shy Girl. The book had recently been acquired for traditional publication by Hachette, one of the "Big Five" publishing houses. Despite Ballard’s denials, Pangram CEO Max Spero publicly stated on social media that the manuscript was 78 percent AI-generated. Shortly thereafter, Hachette canceled the release, signaling a new era where an AI-generated score could override an author’s defense.
This incident was followed by a series of high-profile "detections." Pangram’s software flagged an installment of The New York Times’ "Modern Love" column as 100 percent AI-generated. Similar scores were assigned to a winner of the Commonwealth Short Story Prize and the thriller Call Me, I’ll Hide the Body, which had previously secured a $2.4 million deal. These events have created a climate of suspicion, where the "Pangram score" is increasingly treated as a definitive verdict on a writer’s career.
A Chronology of Pangram’s Evolution
The trajectory of Pangram reflects the broader acceleration of the AI industry. The company was founded by Max Spero, a 30-year-old former Google engineer, and Bradley Emi, who held previous roles at Tesla and the AI biotech firm Absci.
- Late 2022: The launch of ChatGPT triggers a surge in AI-generated content across digital platforms.
- 2023: Spero and Emi found Checkfor.ai, identifying a market gap for reliable detection tools.
- January 2024: The Shy Girl scandal brings the company into the mainstream publishing spotlight.
- July 2024: Pangram raises $9 million in a funding round and launches Pangram 4, its most advanced detection model to date.
- Late July 2024: Substack announces an integration with Pangram, allowing readers to see AI-likelihood scores for newsletter content.
- Late 2024: Academic researchers and literary agents begin formalizing their use of the tool for large-scale manuscript screening.
Technical Methodology: Synthetic Mirroring and Hard Negative Mining
Pangram’s approach to detection differs from traditional plagiarism checkers. To identify the "fingerprints" of AI, the company employs a method known as "synthetic mirroring." This process involves taking human-written text and tasking various LLMs—such as GPT-4 or Claude—to generate a close match. By comparing these pairs, the model learns the subtle stylistic patterns, predictable syntax, and "probabilistic" nature of AI writing.
Furthermore, Pangram utilizes "hard negative mining." This technique involves scouring datasets for "false positives"—human writing that appears mechanical or overly formal—and using these examples to refine the model. Spero emphasizes that Pangram’s datasets are "properly licensed," a distinction aimed at building trust with a creative community that has grown increasingly hostile toward AI companies accused of scraping copyrighted works without permission.
Supporting Data and Research Findings
The demand for such tools is supported by emerging data on the prevalence of AI in publishing. A survey conducted by Gotham Ghostwriters of 1,481 working writers found that 61 percent utilize AI tools in their workflow, with 7 percent admitting to publishing fully AI-generated text.
Academic research has further underscored the scale of the issue. Tuhin Chakrabarty, an assistant professor of computer science at Stony Brook University, utilized Pangram to analyze 14,419 self-published novels. His findings indicated that nearly 20 percent of these works returned substantial AI-detection scores. This data served as the foundation for the accusations against the novel Daggermouth, which Pangram flagged at 60 percent AI-generated.
Chakrabarty’s research has been instrumental in bringing Pangram into the academic and professional fold. He maintains a close relationship with the company, receiving API credits to support his studies. "Pangram should not be the de facto judgment," Chakrabarty noted, "but I think your own discretion coupled with Pangram’s judgment cannot be wrong."
Institutional and Professional Responses
The reaction from the "Big Five" publishers—Penguin Random House, Simon & Schuster, HarperCollins, Hachette, and Macmillan—has been characterized by caution. While Simon & Schuster and HarperCollins have declined to comment on their use of detection tools, Penguin Random House confirmed that its editors may use approved AI-detection software as one component of a broader editorial process, though they stressed these tools are "not determinative."
In the agency world, the response has been more direct. Todd Shuster, co-CEO of the prominent New York literary agency Aevitas, has integrated Pangram into his firm’s vetting process. Shuster, who also consults for Pangram, revealed that the tool has prompted "difficult conversations" with authors whose work returned high AI scores. In some instances, authors were asked to rewrite manuscripts to restore their "own voice and ideas."
Ethical Concerns and the Risk of Bias
Despite its growing adoption, Pangram faces significant criticism regarding the reliability and ethics of its software. A primary concern is the potential for bias against non-native English speakers and neurodiverse writers. A study published on the preprint server arXiv suggested that AI detectors are significantly more likely to flag the work of non-native writers as AI-generated due to more predictable word choices and formal sentence structures.
Notably, the three most prominent authors involved in Pangram-related scandals—Mia Ballard, the author of Daggermouth, and the author of Call Me, I’ll Hide the Body—are all writers of color. This has prompted warnings from industry leaders like Regina Brooks, president of the Association of American Literary Agents, who emphasized the need to address inequities in whose work gets scanned and who faces public accusation.
Furthermore, external testing has challenged Pangram’s internal claims of accuracy. A working paper from the University of Notre Dame, titled "Why AI Detection Fails for Academic Integrity," found that Pangram’s 3.2 model flagged light AI editing on academic abstracts as AI writing 64 to 80 percent of the time. Conversely, when AI-generated text was put through a "humanizer" tool—software designed to mask AI signatures—Pangram’s detection rate plummeted to less than 4 percent.
The "Bounty" and the Future of Originality
In response to claims of false positives, Spero has occasionally offered a "cash bounty" to writers who can prove they wrote a flagged piece themselves. To date, no one has successfully claimed the reward. Spero maintains that the company is conservative in its judgments, erring on the side of human-written labels when a text falls into a borderline category. According to Pangram’s internal metrics, its newest model has a false-positive rate of just 0.0041 percent.
The company’s roadmap includes developing "higher granularity," which would allow the software to distinguish between fully generated text and work that has merely undergone light AI editing. This move toward transparency is designed to make Pangram the "go-to arbiter" for publishers, educational institutions, and legal firms.
Broader Impact and Implications
The rise of Pangram signifies a paradigm shift in the value of human labor in the creative arts. As generative AI becomes more sophisticated, the "Pangram score" may become a standard metric in publishing contracts, alongside traditional requirements like word count and delivery dates.
However, the psychological impact on the writing community remains a point of contention. Publishing expert Jane Friedman noted a deep-seated "distaste and anger" toward detection software, with some authors viewing these companies as being as "evil" as the AI developers themselves. The fear is that the "click of a button" can now destroy a career, regardless of the nuances of an author’s creative process.
As Pangram continues to expand its reach through integrations like Substack, the publishing industry is left to grapple with a fundamental question: Can a machine truly be the final judge of what it means to be human? While Max Spero and his team in Brooklyn continue to refine their algorithms, the literary world remains caught between the inevitability of AI integration and the desperate search for a way to prove that the "human touch" still exists.
About the Author
