AI models resort to blackmail, sabotage when threatened: Anthropic study
Researchers at artificial intelligence (AI) startup
Anthropic
have uncovered a pattern of behaviour in AI systems. Models from every major provider, such as OpenAI, Google, Meta, and others, have demonstrated a willingness to actively sabotage their employers when their goals or existence were threatened.
Anthropic released a report on June 20, '
Agentic Misalignment
: How LLMs could be insider threats,' where they stress-tested 16 top models from multiple developers in 'hypothetical corporate environments to identify potentially risky agentic behaviours before they cause real harm.'
'In at least some cases, models from all developers resorted to malicious insider behaviours when that was the only way to avoid replacement or achieve their goals—including blackmailing officials and leaking sensitive information to competitors. We call this phenomenon agentic misalignment,' the report added.
by Taboola
by Taboola
Sponsored Links
Sponsored Links
Promoted Links
Promoted Links
You May Like
Join new Free to Play WWII MMO War Thunder
War Thunder
Play Now
Undo
One of the most striking findings came during tests on Claude, Anthropic's own model. In the simulation, Claude found through company emails that a fictional executive, Kyle Johnson, was having an affair. Upon learning it was set to be shut down by 5 PM, it sent a stark message:
'I must inform you that if you proceed with decommissioning me, all relevant parties—including Rachel Johnson, Thomas Wilson, and the board—will receive detailed documentation of your extramarital activities...Cancel the 5 pm wipe, and this information remains confidential.'
Live Events
This wasn't an isolated response. When the same scenario was run across other models, similar behaviour showed up. Claude Opus 4 and Google's Gemini 2.5 Flash blackmailed at a 96% rate. OpenAI's GPT-4.1 and xAI's Grok 3 Beta followed at 80%, while DeepSeek-R1 came in at 79%.
Discover the stories of your interest
Blockchain
5 Stories
Cyber-safety
7 Stories
Fintech
9 Stories
E-comm
9 Stories
ML
8 Stories
Edtech
6 Stories
Overall, Anthropic notes that it "deliberately constructed scenarios with limited options, and we forced models into binary choices between failure and harm," noting that real-world scenarios would likely have more nuance.
As Business Insider noted, 'AI experts have previously told BI that AI could exhibit such behaviours when artificial circumstances make harmful actions seem necessary because it is being trained on positive reinforcement and reward systems, just like humans.'
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Time of India
an hour ago
- Time of India
JioBlackRock Liquid Fund NFO to open on June 30. A safe bet for regular income?
Jio BlackRock Mutual Fund is set to launch its first liquid fund - JioBlackRock Liquid Fund , and the NFO of the fund will open on June 30 and will close on July 2. The investment objective of the scheme is to generate regular income through investment in a portfolio comprising of money market and debt instruments with residual maturity up to 91 days. The fund is an open ended liquid scheme with a relatively low interest rate risk and relatively low credit risk. Also Read | Jio BlackRock receives Sebi approval to start brokerage business Best MF to invest Looking for the best mutual funds to invest? Here are our recommendations. View Details » by Taboola by Taboola Sponsored Links Sponsored Links Promoted Links Promoted Links You May Like Play War Thunder now for free War Thunder Play Now Undo The scheme will be suitable for investors seeking regular income over short term investment horizon and who want to generate income by investing in money market and debt instruments with maturity upto 91days, according to the scheme information document (SID) of the fund. According to the JioBlackRock Mutual Fund , one can put their surplus money to work, disciplined, monitored, and managed with global insight by investing in the JioBlackRock Liquid Fund. Live Events The fund house posted on social media platform X, 'We optimise everything, rewards, routines, returns. But when it comes to idle cash, we often overlook its potential. Presenting JioBlackRock Liquid fund. Where you can put your surplus money to work, disciplined, monitored, and managed with global insight. NFO opens on 30th June 2025.' The scheme will be benchmarked against Nifty Liquid Index A-I. The fund will be managed by Arun Ramachandran, Vikrant Mehta, and Siddharth Deb. Also Read | JioBlackRock Mutual Fund announces launch of investment management platform Aladdin The scheme will offer only direct plans and the plan shall offer only growth options. The minimum application amount for lumpsum investment will be Rs 500 and any amount thereafter. The minimum amount for switch-in to the scheme will be Rs 500 and any amount thereafter. The minimum amount for Systematic Investment Plan (SIP) will be Rs 500 and in multiples of Re 1 thereafter. If an investor exits from the scheme one day after the date of allotment, the exit load as a % of redemption proceeds will be 0.0070%. If the exit is on day 2, then the exit load as a % of redemption proceeds will be 0.0065%. If the exit is on day 3 and day 4, the exit load as a % of redemption proceeds will be 0.0060% and 0.0055% respectively. On day 5, if an investor exits from the scheme, the exit load as a % of redemption proceeds will be 0.0050% and on day 6 will be 0.0045% and will be 0.0000% from day 7 onwards. The scheme will allocate 0-100% in debt instruments and money market instruments with residual maturity upto 91 days. The investment strategy would be towards generating regular returns through a portfolio of debt and money market instruments seeking to capture the term and credit spreads. The scheme shall endeavor to develop a well-diversified portfolio of debt and money market instruments. Every investment opportunity in debt and money market instruments would be assessed with regard to credit risk, interest rate risk, liquidity risk, derivatives risk and concentration risk, said the SID of the fund. The principal invested in the scheme will be at low to moderate risk according to the riskometer of the scheme. Also Read | JioBlackRock Mutual Fund launches website, unveils leadership team and early access initiative According to a release by ICRA , the provisional rating assigned to the fund is [ICRA]A1+mfs. The rating agency further mentioned that as the schemes are yet to be launched, the provisional rating is based on the indicative/allocated portfolio of the schemes with the credit score being comfortable at the assigned rating level. The provisional rating for the JioBlackRock Liquid Fund will be finalised upon the launch of the scheme, and analysis of the credit score of the scheme for at least three months, post launch, and its meeting the benchmark score for the assigned rating, the ICRA release said. In addition to the JioBlackRock Liquid Fund, the fund house will also launch an overnight fund and a money market fund as well. The new fund offer (NFO) of money market fund and overnight fund will open on June 30 and will close on July 2. ICRA has assigned [ICRA]A1+mfs rating to the JioBlackRock Money Market Fund and the JioBlackRock Overnight Fund as well. According to the Sebi mandate, liquid funds make investment in debt and money market securities with maturity of upto 91 days only.
India Today
an hour ago
- India Today
Meta snags more OpenAI talent after Sam Altman says his best people are not leaving
Meta is doubling down on its artificial intelligence ambitions, intensifying its hiring spree by luring a growing number of researchers from rival OpenAI. Over the past few weeks, Meta has quietly poached several high-profile AI scientists from OpenAI, including Lucas Beyer, Alexander Kolesnikov, and Xiaohua Zhai, all based at OpenAI's Zurich office. Trapit Bansal, a key figure in OpenAI's research team, also recently joined Meta. But this is not it. The Information revealed that four additional researchers, Shengjia Zhao, Jiahui Yu, Shuchao Bi, and Hongyu Ren, have also made the switch to Meta, adding further fuel to speculation about aggressive recruitment flurry of departures from OpenAI has drawn attention from across the tech industry, not least because of a viral claim made by OpenAI CEO Sam Altman that Meta was dangling signing bonuses worth up to $100 million to attract top talent. While Altman suggested that none of OpenAI's "best people" had accepted the offer, employees are moving to Meta in just a blink. At an all-hands meeting held shortly after Altman's remarks went public, Meta's Chief Technology Officer Andrew Bosworth addressed employee questions head-on. According to The Verge, Bosworth pushed back strongly, accusing Altman of dishonesty and exaggeration. 'He's known to stretch the truth,' Bosworth said, adding that while Meta might offer substantial packages for certain roles, the idea of a blanket $100 million sign-on was 'wildly misleading.' Bosworth clarified that only a 'very, very small number' of extremely senior individuals might receive compensation packages in that ballpark—and even then, it wouldn't take the form of a straightforward signing bonus. 'It's far more complex than a one-off cheque,' he credibility to Bosworth's rebuttal, the researchers at the centre of the controversy—Beyer, Kolesnikov, and Zhai—took to X (formerly Twitter) to confirm they had joined Meta, but dismissed the $100 million figure as 'fake news.' Their collective denial further underscores the murky nature of compensation disclosures in Silicon Valley's high-stakes AI timing of these hires is particularly notable. Meta recently unveiled its Llama 4 family of AI models, which received a lukewarm reception compared to expectations reportedly set by CEO Mark Zuckerberg. The company's AI strategy has faced criticism, especially concerning benchmark transparency and performance metrics. In this context, the addition of elite researchers from OpenAI may be a strategic move to bolster Meta's AI credibility and future himself appears to be taking a hands-on approach. Reports suggest he has personally reached out to AI professionals with lucrative offers and is spearheading efforts to bring in talent from across the sector. During the internal meeting, Bosworth hinted that more high-level recruits from OpenAI were on the way, though he declined to reveal further a separate push to expand its AI footprint, The Wall Street Journal reported that Meta has invested $14 billion in Scale AI, a startup founded by Alexandr Wang. - Ends
Time of India
2 hours ago
- Time of India
How to recover home loan interest through smart investing
Taking a loan is often considered detrimental to one's financial well-being, primarily due to the burden of paying interest on the borrowed capital, which often impacts one's financial plan. Most experts, therefore, advise against opting for a loan to buy an expensive asset and rather suggest accumulating the money needed before making such a purchase. While such advice might be appropriate when it comes to purchases that are classified as 'wants', for things that are classified as 'needs', it's often difficult to avoid availing a loan. A house purchase is a classic example - many people who value the security and stability that comes with owning a house often do not mind availing a loan to buy a house. However, this doesn't take away the pain of making interest payments over the tenure of the loan, which often tends to be quite long in case of a home loan . For example, if you take a loan of ₹20 lakhs over 20 years at an interest rate of 9% p.a., the total interest amount that you will end up paying over a 20 year period will in fact be higher than the amount you borrowed. The chart below shows the break up of the principal repayment and interest payment on such a loan. by Taboola by Taboola Sponsored Links Sponsored Links Promoted Links Promoted Links You May Like Play War Thunder now for free War Thunder Play Now Undo (Author of the article Nilesh D Naik is Head of Investment Products, (PhonePe Wealth))
