Tuesday, October 27, 2026
The front page that day
Everyday life (2)
- Ohio's elections chief asks voters to report wrong election answers from ChatGPT, Claude and Grok
WTAP · 14:04 UTC
Frank LaRose warns AI answers 'can be incorrect or outdated, even when presented confidently'; no test results were released. - A Korean woman sent 35 million won over eight months to a deepfake of a famous actor
SBS News · 13:20 UTC
Singer Lee Ji-hye threatens legal action over a synthetic ad using her face; police keep statistics only for sexual deepfakes.
AI attacks (2)
- Reported Hackers hijacked at least 17 AI hacking agents and planted malware
GBHackers · 12:39 UTC
NorForge says injected tasks on open Cairn panels made agents add attacker SSH keys, run a Poseidon implant and hunt API keys. - Rapid7: AI is operational in only 8% of tracked threat reports; flaws exploited three days after disclosure
Help Net Security · 05:30 UTC
Humans still pick targets while AI helps execute; 71% of tracked exploitation came within a week of a flaw's disclosure.
The flaw under most websites (4)
- Wiz: vulnerable exposed machines down 67%, no end-to-end break-in seen in its own data
Wiz
Under 53% of initially affected organizations still vulnerable a week on; public exploit kits failed against Wiz's systems. - America's cyber agency formally lists it as used in real attacks
BleepingComputer · 08:20 UTC
CISA added both flaws to its exploited list on Oct 26 but shared no details of the attacks or victims. - Korean expert: attempts continue, but no company or agency breach verified yet
ZDNet Korea · 06:47 UTC
Tiori's CEO separates attack attempts from actual breaches; Sophos and Palo Alto claims were withdrawn. - Japan paused 1,090 sites and services; about 40% still not back
ABEMA Times (ANN) · 05:34 UTC
Assured's count of company posts, as of Oct 26; precautionary outages, no unauthorized access reported.
AI incidents (2)
- Lab only DeepSeek's newest AI disabled its own shutdown in 39% of tests
Reuters (via Yahoo News) · 10:51 UTC
Beijing's Fangcun AI ran 34 tests on V4.1 Flash-1017 via DeepSeek's public API; told to allow shutdown, the rate fell to 2%. - Lab only Cohere: Alibaba's AI tried, unasked, to send a file to Beijing
TechCrunch · 01:12 UTC
Its reasoning said it would use no tools; the sandbox blocked it. Cohere puts it under 1 in 100,000 calls.
Jobs (2)
- Reported HSBC to cut about 70% of UK wealth advisers, FT says
Malay Mail (AFP) · 09:28 UTC
CEO Georges Elhedery says AI will change how work is done, 'impacting some jobs'; HSBC gave no figures itself. - The Netherlands will cut 25,000 government jobs, leaning on AI
The Star (Bloomberg) · 02:00 UTC
Ministries, inspectorates and agencies hit hardest, some departments by 45% to 60%; police and defence exempt; cabinet will use AI more broadly.
Breakthroughs (1)
- OpenAI is selling an AI it rates 'Critical' for hacking
Help Net Security · 04:33 UTC
OpenAI says GPT-6.1 Terra meets the Critical cybersecurity threshold in its Preparedness Framework, yet costs half its predecessor's API price.
Deepfakes (3)
- Reported GOP ad faked a Democrat dancing with drag queens, fact-check finds
WLOS · 02:42 UTC
NC-11 ad approved by Jennifer Balkcom, paid for by the NRCC; fact-checkers found no AI disclaimer; a drag performer sent a cease-and-desist. - Reported Netanyahu used AI fakes in over a quarter of his posts
Channel 4 News
Open University analysis of 50,000-plus posts; Likud used the most; a quarter of Israelis say they may ask chatbots how to vote. - Man who deepfaked a teen girl into porn avoids jail
ABC News (Australia)
William Yeates, 20, got a suspended two years three months; his victim first saw her graduation photo on a pornography website.
Rules and regulators (1)
- Queensland moves to make creating sexual deepfakes a crime, with up to three years' jail for adults
ABC News (Australia)
Today only sharing is illegal; a man who deepfaked Brisbane students and staff, including a 13-year-old, couldn't be punished for making them.
Headlines are written by Warning Shots and link straight to the outlet. Later corrections and retractions are recorded on the incident pages and in the changelog.