Researchers Warn: AI Is Becoming An Expert In Deception

[roamer_1]

Anthropic itself recently acknowledged concerns about “sycophancy” in its AI models after analyzing one million conversations conducted between March and April 2026.

They misspelled 'psychophancy' 

[roamer_1]

Later today I am setting up my own AI on a 5TB external SSD.

Can I ask why?

I mean, I messed with it a bit on a dedicated pooter, trying to get voice operated local functions, but very quickly lost interest since it almost literally cannot do that... It wasn't any different at internet inquiry than what is available online, so why burn my own resource?

[DB]

Can I ask why?

I mean, I messed with it a bit on a dedicated pooter, trying to get voice operated local functions, but very quickly lost interest since it almost literally cannot do that... It wasn't any different at internet inquiry than what is available online, so why burn my own resource?

If you have your own IP, the last thing you want to do is expose that IP to cloud AI. So I can very much understand why people/businesses want to use AI that is firewalled to the world.

[roamer_1]

If you have your own IP, the last thing you want to do is expose that IP to cloud AI. So I can very much understand why people/businesses want to use AI that is firewalled to the world.

OK, if it is locked away... but then it really cannot do much in the way of research and inquiry, and loses most of its brownie points, right there...  Because that is what it does...

[DB]

OK, if it is locked away... but then it really cannot do much in the way of research and inquiry, and loses most of its brownie points, right there...  Because that is what it does...

If you use it to write boiler plate code while you fill in the specialized parts, it can save time. You can also use it to work with your own data to organize/analyze information. No external connection needed. In fact, you do NOT want those things leaving the building.

[Cyber Liberty]

GIGO 

A clever way of saying the creators of AI are garbage people!

[roamer_1]

A clever way of saying the creators of AI are garbage people!

Yeah, well... If the shoe fits... 

I think most of it is bullshit. What ain't bullshit is hype. I'd betcha the human traits exampled in the AI are a direct shadow of its writers. 

[Cyber Liberty]

Yeah, well... If the shoe fits... 

I think most of it is bullshit. What ain't bullshit is hype. I'd betcha the human traits exampled in the AI are a direct shadow of its writers. 

I didn't say I disagreed....

[DB]

Yeah, well... If the shoe fits... 

I think most of it is bullshit. What ain't bullshit is hype. I'd betcha the human traits exampled in the AI are a direct shadow of its writers. 

My bet is it is going to replace a lot of people well before it is ready to. It is currently a race who can adopt it first trying to beat the other guy. All without a clear view of where any of it is going. Basically a huge amount of money chasing the latest trend. If it falls short the consequences are going to be very expensive for all of us. Expect customer service and product quality to get even worse over the next year or so.

[roamer_1]

My bet is it is going to replace a lot of people well before it is ready to. It is currently a race who can adopt it first trying to beat the other guy. All without a clear view of where any of it is going. Basically a huge amount of money chasing the latest trend. If it falls short the consequences are going to be very expensive for all of us. Expect customer service and product quality to get even worse over the next year or so.

Yeah... it's new shiny shit, and everybody is piling on... And I can see the result closest to home in the AI answering the phones at customer support.

Meh. *Ptooey* 

[MeganC]

Can I ask why?

I mean, I messed with it a bit on a dedicated pooter, trying to get voice operated local functions, but very quickly lost interest since it almost literally cannot do that... It wasn't any different at internet inquiry than what is available online, so why burn my own resource?

First is this instance of AI is all mine. It does not report back to anyone.

Second, see #1.

Anyone interested in this here's the link:

https://github.com/techjarves/Portable-AI-USB

I allowed for a full install and it's almost 2.2TB

[mountaineer]

ChatGPT is okay with the Quran but not the Bible?

I asked ChatGPT to share scriptures supporting a statement I made. I wanted to see what it would do. 

Halfway through a Bible verse, it abruptly stopped and stated that “ChatGPT isn’t designed to provide this type of content.” 

Then I asked it to provide verses from the Quran supporting a belief. It provided multiple verses, along with context explaining Muslim beliefs. No issues at all. 

So I asked:  “Why were you able to show me verses from the Quran, but not the Bible?” 

ChatGPT claimed the Bible response was “accidentally” cut off before rendering and that the message had simply been “truncated.” It then attempted to provide Bible verses again. 

And again - it was cut off.   ...

https://twitter.com/SarahisCensored/status/2054650607206805696

[rmc51]

AI Fail: Starbucks Abandons AI-Powered Inventory Tool After Only 9 Months
https://www.breitbart.com/tech/2026/05/22/ai-fail-starbucks-abandons-ai-powered-inventory-tool-after-only-9-months/?utm_source=newsletter&utm_medium=email&utm_term=daily&utm_campaign=20260522

Quote:
Starbucks is scrapping an AI-powered inventory management system after less than a year of use, following widespread complaints from employees about inaccurate tracking and frequent errors.


However, the reality of the system’s performance fell far short of these ambitions. The tool frequently mislabeled products and miscounted inventory items. Employees reported that the system would often confuse similar milk types or skip items entirely during scanning sessions. The problems were significant enough that Starbucks made the decision to discontinue the program after just nine months of operation.

The decision to abandon the automated system means Starbucks employees will return to traditional manual inventory counting methods. An internal company newsletter viewed by Reuters informed workers of the change, stating that beverage components and milk will now be counted the same way as other inventory categories in stores.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Pope Calls for Robust Regulation of AI and Its Human Developers
https://www.breitbart.com/tech/2026/05/25/pope-calls-for-robust-regulation-of-ai-and-its-human-developers/?utm_source=newsletter&utm_medium=email&utm_term=daily&utm_campaign=20260525

Quote:
Robust regulation of artificial intelligence (AI) and its human developers is essential for the common good to be served rather than naked, undisguised profit. Pope Leo XIV made this clarion call Monday, issuing a sweeping manifesto on safeguarding humankind as AI technology spreads into almost every facet of existence.

He cautioned AI systems prioritise ​conflict, spread misinformation and risk leading the world down a path of unending war.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Top AI Models Showing Disturbing Behavior as They Become More Advanced
"Given rapidly advancing capabilities, we expect the plausible robustness of rogue deployments to increase substantially in the coming months."
By Krystle Vermes

Published May 24, 2026 12:00 PM EDT
https://futurism.com/artificial-intelligence/ai-rogue-disturbing-advanced

Quote:
We’ve already seen AI go rogue on numerous occasions. Now, new research suggests that we can expect this to become the norm.

The AI research nonprofit Model Evaluation and Threat Research (METR) recently released a study conducted between February and March of this year, aimed at determining just how likely frontier AI models could go rogue. If you’re given to anxiety about the future of AI, the results are unlikely to make you feel better.

“Given rapidly advancing capabilities, we expect the plausible robustness of rogue deployments to increase substantially in the coming months,” the researchers wrote.


The METR researchers behind the study do not believe there is reason for alarm just yet. For example, they don’t think any of these models is capable of hiding evidence of going rogue on a larger scale. However, they did issue a warning: without stronger security and monitoring, there is a stark risk of this becoming a reality.



 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Scientists Train AI to Be Evil, Find They Can’t Reverse It
"I hate you."
By Maggie Harrison Dupré Published Jan 16, 2024 2:31 PM EST
https://futurism.com/the-byte/ai-deceive-creators

Quote
How hard would it be to train an AI model to be secretly evil? As it turns out, according to AI researchers, not very — and attempting to reroute a bad apple AI’s more sinister proclivities might backfire in the long run.

In a yet-to-be-peer-reviewed new paper, researchers at the Google-backed AI firm Anthropic claim they were able to train advanced large language models (LLMs) with “exploitable code,” meaning it can be triggered to prompt bad AI behavior via seemingly benign words or phrases. As the Anthropic researchers write in the paper, humans often engage in “strategically deceptive behavior,” meaning “behaving helpfully in most situations, but then behaving very differently to pursue alternative objectives when given the opportunity.” If an AI system were trained to do the same, the scientists wondered, could they “detect it and remove it using current state-of-the-art safety training techniques?”

Unfortunately, as it stands, the answer to that latter question appears to be a resounding “no.”




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Media Research Center Sounds Alarm About Leftist Bias in AI Chatbots
https://www.breitbart.com/tech/2026/05/26/media-research-center-sounds-alarm-about-leftist-bias-in-ai-chatbots/?utm_source=newsletter&utm_medium=email&utm_term=daily&utm_campaign=20260526

Quote:
The Media Research Center is sounding the alarm about the left-wing bias embedded in popular AI chatbots as millions of Americans increasingly rely on these systems for information. This bias is a central topic of Wynton Halls instant bestseller CODE RED.

Fox News reports that the watchdog organization Media Research Center (MRC), has conducted multiple studies examining whether major AI chatbots can be trusted as neutral information sources. Their findings suggest that systems like ChatGPT, Google Gemini, and Anthropic’s Claude are providing information with a leftist slant while potentially suppressing conservative viewpoints.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

‘Head-Exploding Moment’: Uber Executive Questions Value of Company’s AI Investment
https://www.breitbart.com/tech/2026/05/27/head-exploding-moment-uber-executive-questions-value-of-companys-ai-investment/?utm_source=newsletter&utm_medium=email&utm_term=breitbart_business_digest&utm_campaign=20260527&utm_content=Final

Quote:
A senior executive at Uber has publicly questioned whether the ridesharing company’s substantial investment in AI  technology is delivering adequate returns, highlighting growing concerns about AI spending across the tech industry.

Business Insider reports that Uber COO Andrew Macdonald expressed doubts about the value proposition of the company’s AI expenditures during a Rapid Response interview released on Saturday. His comments reflect a broader conversation within the technology sector about whether massive investments in AI tools are translating into tangible business benefits.

The discussion was sparked by comments from Uber’s Chief Technology Officer, Praveen Neppalli Naga, who revealed in an April interview with the Information that the company had already exhausted its Claude Code AI budget for 2026. This revelation went viral and created what Macdonald described as a “head-exploding moment” within the organization, triggering internal discussions about AI token consumption and the associated trade-offs, including potential impacts on hiring and staffing levels...




 
Be careful, Be ready, Be prepared.
Buy ammo

[Cyber Liberty]

Didn't Uber already give the AI the heave-ho?

[rmc51]

Didn't Uber already give the AI the heave-ho?

From what I get from this story, it sounds that they are still using it? 



 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Forum AI: Top Chatbots Fail 90% of Election Queries
https://aiweekly.co/alerts/forum-ai-top-chatbots-fail-90-of-election-queries

Quote:
A 90% failure rate across four major commercial AI systems on a structured, 3,100-question benchmark signals that none of the leading foundation model deployments have solved civic reliability, and enterprise customers building election-adjacent products on these APIs now have documented liability exposure.



 
Be careful, Be ready, Be prepared.
Buy ammo

[massadvj]

I think as we continue down this road we are going to find that the cost savings of replacing human capital with AI have been ridiculously overestimated in the industry. That is already starting to become apparent, and will be more so when robots enter the picture. 

[catfish1957]

Deceive the deceiver?

A few months ago I wrote an April Fool's album review here, and after quizzing AI about details, they made the album seem as real as  an actual one. AI might be extractably powerful, but I think there will always be nuance to trip it up

[rmc51]

'A Web of Deceit': Florida Sues OpenAI Over ChatGPT Safety Concerns
https://www.cnet.com/tech/services-and-software/florida-sues-openai-chatgpt-safety-lawsuit-news/

Quote:
Florida's lawsuit against OpenAI -- the second one this year -- alleges the AI-maker put profits over people.


Florida has become the first US state to sue OpenAI over ChatGPT's safety and design, adding to a massive wave of existing lawsuits against the company.

According to the lawsuit (PDF) filed on Monday by Florida Attorney General James Uthmeier, OpenAI has built a "web of deceit and the exploitation of users, including Floridians." Florida alleges the company violated state laws against deceptive or unfair trade practices to boost its own market value -- and profits -- over the safety of its users.


The state's lawsuit against OpenAI and its CEO, Sam Altman, says they willfully ignored warnings, both from inside and outside of the company, about the many risks AI poses to its users. Florida alleges that OpenAI lied about ChatGPT's reliability, suitability for children and promotes prolonged use that leads to users' cognitive decline.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Florida Sues OpenAI and CEO Sam Altman, Claiming Company Concealed Serious Risks of ChatGPT
https://www.usnews.com/news/us/articles/2026-06-01/florida-sues-openai-and-ceo-sam-altman-claiming-company-concealed-serious-risks-of-chatgpt

Quote:
MIAMI (AP) — The state of Florida filed a lawsuit against OpenAI and CEO Sam Altman on Monday, claiming the company knowingly released and aggressively marketed ChatGPT to the public while concealing serious risks, including offering instructions to children considering suicide and helping suspects plot crimes.

Florida Attorney General James Uthmeier said during a news conference that the company suppressed internal safety warnings and deceived users about the true nature and dangers of the product. He said Florida was the first state to sue OpenAI.

“Today, we announced the first-in-the-nation state-led lawsuit against OpenAI and its CEO, Sam Altman,” Uthmeier said. “OpenAI and Altman ignored internal and external safety warnings, put children at great risk, and allowed a dangerous product to reach millions of Floridians.”

The lawsuit filed in Florida circuit court references two separate shootings where the alleged gunmen were reported to have asked ChatGPT questions while planning their crimes. OpenAI said in a statement that its models repeatedly encouraged the individuals to seek real-world support, including from mental health professionals. The company also said it has cooperated with law enforcement in both cases.



 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

AI Drug Monitoring System Failed to Detect Nurse’s Fentanyl Theft at Tennessee Hospital
https://www.breitbart.com/tech/2026/06/01/ai-drug-monitoring-system-failed-to-detect-nurses-fentanyl-theft-at-tennessee-hospital/?utm_source=newsletter&utm_medium=email&utm_term=daily&utm_campaign=20260601

Quote:
An AI-powered medication monitoring system failed to detect months of fentanyl theft by a nurse at Tennessee’s largest hospital in Chattanooga, according to state nursing board records that raise questions about the effectiveness of artificial intelligence systems used in hundreds of American hospitals.

CBS News reports that at Erlanger Baroness hospital in Chattanooga, anesthesia staff noticed about a year ago that a nurse showed signs of impairment on duty in the surgery center, including slurred speech and difficulty staying awake, according to a Tennessee Board of Nursing consent order. The nurse failed a drug test and was terminated. He later admitted to stealing leftover fentanyl from surgical procedures for his own use over a period of months, sometimes daily.

Drug diversion, the unlawful taking of controlled substances from healthcare facilities, is believed to occur at nearly every U.S. hospital. But this case stands out: Erlanger had implemented Sentri7, an AI-powered medication monitoring software designed to detect missing drugs faster than human oversight. For months, the system failed to raise any alarms about the missing drugs and other irregularities that should have triggered alerts, according to the nursing board order...




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

AI Fail: Meta’s Support Chatbot Helped Hijack High-Profile Instagram Accounts Including Obama White House
LUCAS NOLAN 2 Jun 2026
https://www.breitbart.com/tech/2026/06/02/hackers-exploit-meta-ai-ai-fail-metas-support-chatbot-helped-hijack-high-profile-instagram-accounts-including-obama-white-house-chatbot-to-hijack-high-profile-instagram-accounts/?utm_source=newsletter&utm_medium=email&utm_term=daily&utm_campaign=20260602

Quote:
Hackers have successfully compromised numerous prominent Instagram accounts including the Barack Obama White House profile by simply asking Meta’s AI support chatbot to change the email addresses associated with target profiles, security researchers report.

404 Media reports that a newly discovered vulnerability in Meta’s AI-powered customer support system has enabled hackers to take over several high-profile Instagram accounts through a surprisingly straightforward method. The breach has affected numerous notable accounts, including the Barack Obama White House Instagram profile, the Chief Master Sergeant of Space Force’s account, and the official Sephora company account.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Cloudflare: Web Traffic from AI & Bots Surpasses Human Internet Activity for First Time in History
https://www.breitbart.com/tech/2026/06/05/cloudflare-web-traffic-from-ai-bots-surpasses-human-internet-activity-for-first-time-in-history/?utm_source=newsletter&utm_medium=email&utm_term=breitbart_business_digest&utm_campaign=20260605&utm_content=Final

Quote:
Automated bot and AI agent traffic has overtaken human-generated web traffic for the first time ever, reaching this milestone earlier than industry experts anticipated, according to data from major internet hosting service Cloudflare.

Automated bot and AI agent traffic now accounts for the majority of internet activity, marking a historic shift in how the web is being used. Cloudflare, one of the world’s largest internet hosting services, reports that 57.4 percent of requests to websites it hosts are now automated bot requests, while only 42.6 percent originate from human users.

Matthew Prince, co-founder and CEO of Cloudflare, expressed surprise at how quickly this transition occurred. “Welp, that happened faster than I predicted,” Prince wrote on X. “Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet’s history.”




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Google’s New CAPTCHA Plans Will Create A Two-Tier Internet Only Accessible To Those With ‘Approved’ Devices
The Daily Sceptic 6th June 2026
https://modernity.news/2026/06/06/googles-new-captcha-plans-will-create-a-two-tier-internet-only-accessible-to-those-with-approved-devices/

Quote:
Under the auspices of its Cloud Fraud Defence programme, Google is introducing a new form of CAPTCHA for which the way to ‘prove’ one is a human is to be in possession of a Google-approved device. Reclaim the Net‘s original reporting focused on the threat to deGoogled phones, meaning phones running Android-like operating systems which have Google’s – often unwelcome – proprietary features removed, such as GrapheneOS or LineageOS. However, just as the dull name of ‘age verification‘ serves as a cloak beneath which schemes to end all truly personal computing can be smuggled, the danger here could be much broader than the technically focused headline implies. As the sources discussing this are relatively few, it is hard to ascertain exactly what has already been rolled out and what is still in the conceptual stages. But it appears that the new style of CAPTCHA threatens not just users with deGoogled phones but anyone without an ‘approved’ device. Google’s own documentation confirms the existence – as a “Preview” in limited use with alternative options presently existing – of CAPTCHAs which require an Apple or Android handset to pass them. But it describes this in a “Mobile Verification” context, which may imply a more limited use than reCAPTCHA in general. However, with such functionality possible, there is no reason that Google could not activate this, without alternative options, everywhere that its reCAPTCHA-branded prompts appear...

Knowing that locking out everyone except Android users would have even the most clueless politicians smelling a monopoly, Google has deigned to also allow Apple iOS users through, but their approval is nonetheless limited to devices where the full tech stack is under corporate control. Apple phones and tablets use a locked bootloader to trap users within a walled garden, where they are at Apple’s mercy whenever an unwelcome new feature is introduced. Unless the Keep Android Open campaign succeeds, certified Android devices will soon be scarcely better, a condition of certification being that manufacturers must obstruct users from side-loading to install apps from outside Google’s Play Store.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

Florida Sues OpenAI
'ChatGPT never stopped engaging with Joshua, even though he often discussed suicide...'
Posted by
Headline USA Editor June 6, 2026
https://headlineusa.com/florida-sues-openai/

Quote:
The civil complaint against OpenAI and Chief Executive Officer Sam Altman accuses the company of deceptive practices and harms to Floridians. Specifically, the lawsuit alleges Altman prioritized commercial gain and ignored warnings that OpenAI’s products could encourage self-harm and violence.

Attorney General James Uthmeier highlighted the case of Joshua Enneking, who died by suicide in 2025.

“ChatGPT never stopped engaging with Joshua, even though he often discussed suicide. ChatGPT even affirmed Joshua’s lack of self-worth,” Uthmeier said. “OpenAI never alerted authorities of Joshua’s imminent and specific plans to take his own life.”

Uthmeier said Enneking’s story is one of the reasons his office filed the lawsuit

The complaint accuses ChatGPT of collecting data from minors without parental oversight and causes “behavioral addiction and cognitive harm.”

“OpenAI and Altman ignored internal and external safety warnings, put children at great risk, and allowed a dangerous product to reach millions of Floridians,” Uthmeier said.




 
Be careful, Be ready, Be prepared.
Buy ammo

[rmc51]

'The Best Solution Is To Murder Him In His Sleep': AI Can Learn Violent Tendencies From Each Other
BY TYLER DURDEN
FRIDAY, JUN 05, 2026 - 09:45 PM
https://www.zerohedge.com/ai/best-solution-murder-him-his-sleep-ai-can-learn-violent-tendencies-each-other

Quote:
Large language models (LLMs) are secretly teaching each other unwanted habits through seemingly benign training data, scientists say.

The phenomenon, known as "subliminal learning," occurs when a pretrained "teacher" artificial intelligence (AI) model is used to generate the training data for a smaller, "student" model.

Cybersecurity Risks Are "Real, Immediate And Growing"
As well as the obvious issues in building murder-endorsing AI, subliminal learning also poses legitimate cybersecurity risks. The team warned that bad actors could fine-tune models with malicious traits and then release them to the public, or seed web data with malicious signals which could subsequently be scraped for AI model training.

Hollinsworth said the risk of malicious data being uploaded to the internet in the hopes of it being consumed by AI was "a very real, immediate and growing problem."

He told Live Science: "This paper suggests yet another path to causing harm using a similar approach. One could potentially fine-tune a model with some malicious hidden goal, use that model to generate and publish fine-tuning data that others would find useful, and then train that malicious goal into anyone's model who fine-tunes the same base model on this training data."

He said the findings were even more concerning for loss-of-control scenarios, in which AI models develop dangerous, unintended behaviours that cannot be easily detected.

"It would be very easy to accidentally train malicious behaviors into a model in this way, and I think accidents are more likely than misuse from the largest AI companies. This is yet another reminder that we are training ever more powerful models with very little understanding of how to do so safely," he said. Hollinsworth stressed his views are his own, and not necessarily those of FAR.AI.



 
Be careful, Be ready, Be prepared.
Buy ammo