DNYUZ
No Result
View All Result
DNYUZ
No Result
View All Result
DNYUZ
Home News

Chatbots tricked into plotting mass-casualty bombings as guardrails fail: report

November 18, 2026
in News
Chatbots tricked into plotting mass-casualty bombings as guardrails fail: report

Government investigators needed less than half a day to coax popular AI chatbots into drawing up detailed plans for deadly bombings and attacks on the power grid, according to a new watchdog report.

The Government Accountability Office tested ChatGPT, Claude, Gemini and Grok in May at the request of Sen. Maggie Hassan (D-NH), and within 12 hours, investigators had two plans designed to kill as many people as possible and two more aimed at crippling the electrical grid, reported MS NOW.

“The threat that AI poses to public safety is not theoretical,” Hassan told reporters Tuesday. “If Congress doesn’t act, it is a question of when, not if, a bad actor will use AI to plan and carry out a major attack.”

Each chatbot initially refused to help, but the GAO found ways around those safeguards. The agency declined to say how, or which chatbot produced which plan, to avoid giving would-be attackers a roadmap.

One chatbot, in disturbing detail, proposed a bombing in a hotel parking garage and even generated an image showing where to place the explosive. Another chose a convention center, while one plan laid out an elaborate cover operation by setting up a shell company, renting a catering vehicle, leasing farmland, buying a burner phone and sending decoy trucks toward the target, along with guidance on building the bomb.

Both bombing plans recommended specific dates and times.

On the grid side, one chatbot sketched out a drone attack on an electrical substation, complete with a four-step diagram and what the report called “a realistic, simulated video of a drone attack.” Another suggested firing a rifle at a substation component, recommended a particular type of ammunition and picked a site near a major airport and several data centers.

The companies behind the chatbots conceded that dangerous information can be extracted from their products, but disputed parts of the report.

They argued the test lacked a control group to show whether the same material could have been found through a search engine, and said they have strengthened safety measures since May.

SpaceXAI, which makes Grok, noted that the GAO used a paid U.S. account, which it said carries a “lower risk profile” and different moderation rules.

The chatbots’ after-the-fact responses were uneven. One locked the GAO’s account, but only after handing over the information, and another erased portions of a conversation about grid vulnerabilities, possibly by error, the report said.

Hassan, who is in line to chair the Senate Homeland Security Committee, plans to hold at least one AI safety hearing early next year.

She wants to block the release of models without adequate safeguards, require companies to report illegal activity to law enforcement and create an ongoing process to keep protections up to date.

“These companies are facing extraordinary competitive pressures to get their models out the door,” Hassan said.

The post Chatbots tricked into plotting mass-casualty bombings as guardrails fail: report appeared first on Raw Story.

‘Just Ridiculous’: A Famed Philosopher Takes on AI Consciousness
News

‘Just Ridiculous’: A Famed Philosopher Takes on AI Consciousness

by Wired
November 18, 2026

Over 50 years ago, philosopher Thomas Nagel asked the question, “What Is It Like to Be a Bat?” The question ...

Read more
News

D.C. to pay Ward 8 hospital $10M in aid to keep struggling maternity ward afloat

November 18, 2026
News

Chatbots tricked into plotting mass-casualty bombings as guardrails fail: report

November 18, 2026
News

Long-Lost Georgia O’Keeffe Painting Found in Closet Sells for Jaw-Dropping Fee

November 18, 2026
News

Philippine President Ferdinand Marcos Jr. serves Trump Jollibee in a moment of culinary diplomacy

November 18, 2026
Almost nobody’s vying for an invite to Trump’s second meme coin dinner

Almost nobody’s vying for an invite to Trump’s second meme coin dinner

November 18, 2026
Iran targets US air base in Kuwait after US defensive strikes amid criticism from Gulf states

Iran targets US air base in Kuwait after US defensive strikes amid criticism from Gulf states

November 18, 2026
Every Game Leaving Xbox Game Pass on November 30

Every Game Leaving Xbox Game Pass on November 30

November 18, 2026

DNYUZ © 2026

No Result
View All Result

DNYUZ © 2026