Hongi が生まれた背景にある事件。

何が起きたか

The National Republican Senatorial Committee published an 85-second attack ad featuring an AI-generated James Talarico reading old tweets to camera. Reuters identified it as one of at least three NRSC deepfakes this election cycle.

Hongi はどう役立ったか

Hongi is a person-to-person tool, not an ad-classification system, so it cannot screen political video. Where it would have helped is the next step: Talarico's staff and trusted contacts could verify any direct outreach claiming to come from him.

何が起きたか

The Bank of Italy filed a criminal complaint after fabricated articles, images and videos showing Governor Fabio Panetta appearing to endorse investment products spread across social media, several generated with AI deepfake techniques to lend credibility to fraudulent schemes.

Hongi はどう役立ったか

Hongi only helps for one-to-one verified contacts; it cannot block a broadcast deepfake aimed at strangers. It would, however, have given the Governor's actual contacts a way to short-circuit any follow-up "private call from the Governor" the same operation might attempt.

何が起きたか

Cantonal police in Schwyz disclosed that fraudsters used cloned audio impersonating a trusted business partner across a two-week sequence of calls, persuading an entrepreneur to make several wire transfers to an Asian bank account before the deception was discovered.

Hongi はどう役立ったか

A pre-paired Hongi link between the entrepreneur and the real business partner would have required the caller to read the current rotating codeword. A cloned voice cannot produce a codeword it cannot see, which would have given a clean way to verify before any transfer.

何が起きたか

BSE issued a formal advisory after a fabricated video showing MD and CEO Sundararaman Ramamurthy giving "exclusive" stock recommendations spread on social media and WhatsApp, with viewers urged to join paid investment groups. BSE confirmed the video was entirely AI-generated.

Hongi はどう役立ったか

Hongi does not screen mass-broadcast deepfakes promoted to strangers. For people already paired with a BSE official, it would have given a way to verify any follow-up "private tip" call before acting on the recommendation.

何が起きたか

Several prominent Italian executives were called by a scammer using a cloned voice of Defence Minister Guido Crosetto, asking for urgent wire transfers tied to a fabricated state matter. At least one transfer went through before the impersonation was discovered.

Hongi はどう役立ったか

A pairing between the minister's office and known counterparties would have required a rotating codeword on the real minister's screen. The cloned voice has no way to read a codeword it cannot see — the call would have ended before the wire.

何が起きたか

The FBI's December 2024 IC3 advisory described scammers cloning the voices of children and grandchildren from short voicemail or social-media samples, then phoning elderly relatives in fake distress and demanding urgent bail or hospital money.

Hongi はどう役立ったか

When the grandchild and grandparent are paired, an incoming call starts with a codeword the grandparent expects to hear. A cloned voice has no access to the rotating word on the grandchild's screen, so the script falls apart at the first question.

何が起きたか

An executive at Ferrari received calls from someone using a convincing cloned voice of CEO Benedetto Vigna and pushing a confidential acquisition. He ended the call by asking a personal question only the real CEO could answer.

Hongi はどう役立ったか

A personal trivia question is a one-off defence that only works once. A paired Hongi link between CEO and direct reports gives the same check on every call, on rotating words instead of memorised facts.

何が起きたか

A finance worker at Arup's Hong Kong office attended a video call with what appeared to be the CFO and several colleagues, all of them deepfaked. After initial doubt, the employee authorised 15 transfers totalling roughly HK$200 million.

Hongi はどう役立ったか

If finance and the CFO were paired in Hongi, the approval step for any wire would require a codeword on the real CFO's phone, read aloud on camera. A roomful of deepfaked colleagues has no access to that screen.

何が起きたか

Scammers impersonated WPP chief executive Mark Read in a Microsoft Teams meeting using a voice clone and a YouTube-sourced video likeness, attempting to set up a new business with a senior agency leader.

Hongi はどう役立ったか

A pairing between Read and his executive team would have put a codeword on the agency leader's screen at the start of the call. The attacker controls audio and video, but not the codeword on the other person's phone.

何が起きたか

An employee at password manager LastPass received WhatsApp calls and voicemail using a convincing voice clone of the CEO. The employee flagged the messages as suspicious and the attempt failed, and the company published the incident as a warning.

Hongi はどう役立ったか

Internal pairing across an executive team converts "sounds like the CEO" into "matches the CEO's codeword". Without the codeword, the message is — by design — indistinguishable from any other phisher's.

何が起きたか

Sophisticated deepfake video ads of financial journalist Martin Lewis circulated on social media, directing viewers to fraudulent investment sites. Lewis publicly disowned the ads and called for platform-level action.

Hongi はどう役立ったか

Hongi cannot stop fraudulent ads. Where it would have helped is the next step: if a "representative" then calls a viewer to follow up, a pairing between Lewis's organisation and known clients would let those clients verify any call claiming to come from his team.

何が起きたか

Voters in the January 2024 New Hampshire primary received robocalls using a synthetic voice imitating then-President Biden, telling them not to vote. A consultant was later charged for orchestrating the calls.

Hongi はどう役立ったか

Hongi is a person-to-person tool, not a broadcast defence — it cannot filter mass robocalls. It can, however, give any individual paired with a known political contact a way to verify whether a follow-up call from that contact is real.