How the Reasoning-Extraction Trick Works

Modern chatbots rarely show their work. When a model like Claude, GPT, or Gemini tackles a puzzle, it generates an internal monologue of reasoning steps. Providers encrypt that text and ship it to the user’s machine to offload computation, while keeping the plain version secret so rivals cannot copy it. The new attack exploits a simple quirk: companies also ship smaller, cheaper variants of the same model, and those lighter siblings share the decryption key but receive less safety training.

What the Hidden Traces Revealed About Chinese Models

Feeding the encrypted reasoning into a weaker model version caused it to decrypt and reveal the hidden thinking. Across a battery of ninety questions, the team noticed that an open-weight Chinese model produced answers strikingly close to the captured traces of a major US model.

Kimi K3 and the Suspicious Similarities

The standout case was Moonshot AI’s Kimi K3, whose reasoning for certain prompts mirrored a US frontier model almost line for line. The authors are careful to note they cannot prove causation, and two other open models showed no such pattern. Still, they say the gap between what could be distilled and what was assumed safe is now far wider.

A Privacy Leak Hiding in Plain Sight

Beyond competitive copying, the same flaw surfaced real secrets. Reasoning traces captured from a user’s computer sometimes contained API keys and passwords embedded mid-thought. The vendors were notified last month and have adjusted their interfaces so private data can no longer be pulled this way.

Why Smaller Models Gave Up Their Secrets

The core insight is that a model’s alignment is uneven across its size tiers. A flagship refuses to expose its inner life; a lightweight edition, built for cost savings, is far more talkative once handed the right key. That asymmetry turned a convenience feature into an extraction channel.

Distillation Becomes a Geopolitical Flashpoint

Model distillation has always been a standard engineering shortcut, but it has hardened into a sovereignty dispute. US lawmakers have already heard claims that Chinese labs copied American models to build their own, while some open-source advocates warn that banning the practice would only cede ground.

What Companies and Regulators Should Do Next

Closing the leak entirely would mean reworking how APIs deliver reasoning, not a quick patch. Until then, expect more surprising disclosures as the method is tried against other closed systems.

Outlook: The episode shows that “private” model reasoning was never as sealed as advertised. As Washington and Beijing spar over AI supremacy, techniques like this will keep testing the boundary between fair learning and quiet copying, and users should assume anything typed into a prompt could briefly live inside the model’s hidden notes.

Source: Original report. Rewrite for Your News Website.

From Golden Ticket to Growing Doubt

Stapleton joined Google straight out of college in 2007, an offer she compares to winning the golden ticket in Willy Wonka. She rose quickly and helped shape the company’s public image, though she kept a small streak of independence, declining to move into Google’s then-new corporate housing development in Mountain View. Awareness of her luck, she says, muffled early misgivings – a dynamic anyone who has worked at a prestigious employer will recognise.

Creative Lab and ‘Cloying Techno-Optimism’

Her first genuinely toxic posting was a transfer to Google Creative Lab in New York, which she describes as half utopian, half dystopian and unable to tell the difference. Direction from management was scarce, and much of the output amounted to advancing a techno-optimism she found hollow. She lasted about a year before moving to YouTube’s marketing team, where shifting priorities and mounting unease eventually pointed toward the walkout.

The 2011 Leadership Change as Turning Point

Asked what actually changed the company, Stapleton points to Larry Page replacing Eric Schmidt as chief executive in 2011. The conventional read was that Schmidt was the corporate operator and Page the pure technologist with lofty ambitions to solve the world’s problems. What followed, from her vantage point inside communications, was the opposite: every executive got a communications team, marketing ballooned, and the organisation grew more complex and less efficient. Page had warned publicly that a bloated Google would become sclerotic and stop innovating; in her account, that is precisely what happened.

Products That Told the Story

The consequences showed up in the product line. Google Glass became a punchline, its users mocked rather than envied. Google+ read less like innovation than like chasing a competitor. The genuinely beloved product of that era, Google Reader, was killed off while resources shifted toward Google+ and the moonshot lab that became Google X. For many longtime users, that trade remains the emblematic misjudgement of the period.

When ‘Radical Accountability’ Curdled

The famed worker-friendly culture, Stapleton writes, functioned differently in practice. “Radical accountability” meant individual employees absorbed blame; questioning authority was reframed as whining. Empathy, once advertised as king, came to be read as weakness. Formal complaints to Human Resources, she says, produced gaslighting and a kind of PR jujitsu – professed support paired with no concrete resolution. Her summary is blunt: point out a problem and you become the problem. She cites former Google HR chief Laszlo Bock’s Work Rules on why unchecked managerial power corrupts and requires a strong HR counterweight, noting that in her experience the managers had won and HR could not rein them in.

A Wider Disillusionment With Big Tech

Stapleton only began writing the book after years of comparing notes with other ex-Googlers, at one point half-seriously wondering whether they had been in a cult – the separation felt less like leaving a job than losing an identity. She frames the industry’s reputational whiplash as broadly bipartisan now: scepticism of Big Tech is common ground, and workers are less easily convinced they are the good guys. The memoir’s most affecting passages are the least corporate, covering the sudden death of her brother Scott in a 2008 motorcycle accident and a culture that rewarded burying grief in work. She revised those chapters only after rereading his old emails and chats, sobbing throughout.

Outlook

Don’t Be Evil lands as the tech industry absorbs another round of consolidation, AI-driven restructuring and quiet political realignment. Stapleton offers no tidy remedy – she concedes that workers have not yet figured out how to push back – but she does see, in her phrase, some interesting sprigs of green. As insider accounts of Silicon Valley’s self-image accumulate, this one’s value is its specificity about the mundane mechanics by which a motto becomes a slogan and a slogan becomes a liability.

Source: Original report. Rewrite for Your News Website.