Google actively markets artificial intelligence solutions to corporations as a means of rapidly processing vast quantities of job applications and identifying the strongest candidates. Yet within the technology giant itself, some of the company's most respected researchers have expressed serious reservations about these very systems, creating an uncomfortable paradox that highlights the gap between AI marketing promises and operational reality.
DeepMind's AGI Safety and Alignment Team, responsible for studying how to prevent risks from increasingly powerful artificial intelligence systems, has taken an unusual step in its own recruitment drive. The team is asking prospective employees to complete a supplementary form alongside their standard application materials, a workaround designed explicitly to ensure their submissions bypass the company's internal automated screening mechanisms. An internal document marked with a warning against wide circulation acknowledged what the team's leadership clearly believes: their employer's AI hiring systems have a "non-trivial probability" of incorrectly filtering out qualified applicants or substantially delaying their review.
The disclosure creates an awkward situation for Alphabet Inc's Google, which has spent considerable resources convincing enterprise clients that its AI tools represent a superior approach to recruitment management. The company's own researchers—the very people developing cutting-edge AI safety protocols—apparently lack confidence in these screening systems when their own careers are at stake. This contradiction suggests that even within Google's walls, there remains scepticism about whether current AI hiring technology can reliably identify talent without introducing errors.
When questioned about the practice, a Google DeepMind representative denied that the company's systems produce faulty results, instead characterising the special form as simply an alternative pathway for candidates to reach hiring teams directly. The spokesperson maintained that Google DeepMind remains committed to recruiting the highest-calibre talent and that no shortcuts exist in the hiring decision-making process itself. However, this explanation does not fully address the team's own assessment that their applications system contains meaningful risks of incorrect screening.
The practice of using artificial intelligence in hiring has grown substantially across the corporate sector, driven by the efficiency gains such systems promise. Some organisations deploy AI models to rank and score applicants, while others use algorithms to scan résumés for specific keywords and credentials. Google's own Workspace division, which sells productivity tools like Google Drive to businesses worldwide, actively promotes new AI capabilities that ostensibly help human resources departments save time by drafting job postings, evaluating résumés, and forecasting hiring requirements. This represents a significant revenue stream for the company, making the DeepMind team's implicit criticism particularly significant.
Beyond Google, the use of AI in hiring has attracted scrutiny from researchers and regulators concerned about potential discrimination. A recent Bloomberg investigation revealed that OpenAI's ChatGPT exhibited signs of bias when evaluating applicants based on their names, suggesting that seemingly neutral algorithms may reflect or amplify existing societal prejudices. In a separate case with direct legal consequences, Workday Inc, which supplies workplace management software to thousands of organisations globally, faces litigation alleging that its AI hiring systems discriminate against candidates on grounds of race, age, and disability, violating employment law. Workday maintains that human judgment drives ultimate hiring decisions and has declined to comment further on the allegations.
The problem extends in both directions. While companies worry about AI systems making flawed decisions, job seekers have begun exploiting these automated filters themselves. Some candidates now use generative AI tools to rapidly produce applications tailored to multiple positions, dramatically increasing the volume of submissions they can make. Others strategically incorporate keywords and formatting designed to appeal to algorithmic screening. The DeepMind team anticipated this challenge and included a warning in their special form cautioning applicants that human reviewers become fatigued by AI-generated text, which tends to share recognisable stylistic patterns. This guidance suggests that candidates hoping to succeed should present authentic, individually crafted materials rather than algorithmic outputs.
For Malaysian and Southeast Asian readers, this situation carries particular relevance. As artificial intelligence adoption accelerates across the region's corporate sector, many multinational companies and local enterprises are implementing similar AI-driven hiring systems. The revelation that even Google's own safety-focused researchers doubt their company's screening technology should prompt local organisations to carefully evaluate the AI hiring solutions they deploy. The risks of algorithmic bias take on added importance in diverse societies like Malaysia, where employment discrimination based on ethnicity, religion, or other characteristics remains a documented concern. Companies implementing these systems should ensure robust human oversight and regular audits for potential bias.
The contrast between how Google promotes its AI hiring capabilities to clients and how its own teams approach recruitment underscores a broader challenge in the technology industry: the gap between marketing claims and practical limitations. While AI undoubtedly offers genuine efficiency benefits, the DeepMind team's caution suggests that current systems should not replace human judgment in high-stakes decisions like hiring. For job seekers navigating this increasingly AI-mediated employment landscape, the most prudent approach may be exactly what the DeepMind team suggests—ensure that real humans, not algorithms alone, evaluate their qualifications and potential.
