The Japan Times - AI is learning to lie, scheme, and threaten its creators

EUR -
AED 4.040576
AFN 70.414639
ALL 90.713366
AMD 393.670515
ANG 1.969846
AOA 1009.455967
ARS 1705.222388
AUD 1.633374
AWG 1.983155
AZN 1.873245
BAM 1.953669
BBD 2.214732
BDT 136.917515
BGN 1.951216
BHD 0.415005
BIF 3306.175169
BMD 1.100225
BND 1.417493
BOB 12.828121
BRL 6.364692
BSD 1.099575
BTN 103.226383
BWP 15.393206
BYN 3.30422
BYR 21564.403763
BZD 2.211536
CAD 1.572832
CDF 2550.320671
CHF 0.923842
CLF 0.0279
CLP 1098.068061
CNY 7.332502
CNH 7.33193
COP 3651.612712
CRC 490.00974
CUC 1.100225
CUP 26.390211
CVE 110.435001
CZK 24.414539
DJF 195.531637
DKK 7.475047
DOP 66.618208
DZD 150.803335
EGP 59.062154
ERN 16.50337
ETB 177.410959
FJD 2.513738
FKP 0.832383
GBP 0.834295
GEL 2.855082
GGP 0.832383
GHS 12.91116
GIP 0.832383
GMD 80.869244
GNF 9747.990648
GTQ 8.404792
GYD 230.098968
HKD 8.632489
HNL 29.782904
HRK 7.516697
HTG 144.122766
HUF 364.379564
IDR 20101.104936
ILS 3.568744
IMP 0.832383
INR 103.361763
IQD 1666.840393
IRR 2118211.969194
ISK 136.801391
JEP 0.832383
JMD 174.365809
JOD 0.780069
JPY 175.370293
KES 143.084631
KGS 96.213987
KHR 4488.916782
KMF 492.351175
KPW 990.202599
KRW 1470.23575
KWD 0.342852
KYD 0.9165
KZT 489.306179
LAK 24782.560737
LBP 98525.119758
LKR 365.990713
LRD 185.388661
LSL 18.687297
LTL 3.248677
LVL 0.665515
LYD 7.101971
MAD 10.999221
MDL 20.114056
MGA 5054.432079
MKD 61.554773
MMK 2309.99731
MNT 3965.241369
MOP 8.887604
MRU 44.724055
MUR 52.61237
MVR 16.954496
MWK 1911.0899
MXN 20.37928
MYR 4.504321
MZN 70.31547
NAD 18.687339
NGN 1480.7707
NIO 40.433267
NOK 10.623
NPR 165.161813
NZD 2.000131
OMR 0.423038
PAB 1.09971
PEN 3.749782
PGK 4.966442
PHP 70.324708
PKR 304.074063
PLN 4.401724
PYG 6824.399678
QAR 4.009213
RON 5.342361
RSD 117.639325
RUB 90.880609
RWF 1625.031855
SAR 4.140192
SBD 8.910016
SCR 15.249738
SDG 661.796312
SEK 11.10591
SGD 1.418075
SHP 0.83461
SLE 25.360078
SLL 23071.165496
SOS 629.893518
SRD 41.207269
STD 22772.429568
STN 24.617527
SVC 9.622284
SYP 14305.121683
SZL 18.687339
THB 37.283867
TJS 10.133944
TMT 3.861789
TND 3.3469
TOP 2.649077
TRY 55.478391
TTD 7.468004
TWD 36.283199
TZS 2910.097592
UAH 49.988809
UGX 4563.502856
USD 1.100225
UYU 44.650273
UZS 13230.201849
VES 1006.269457
VND 28346.188701
VUV 132.818678
WST 3.092274
XAF 655.957
XAG 0.02337
XAU 0.000285106532
XCD 2.973412
XCG 1.981938
XDR 0.777916
XOF 655.957
XPF 119.331742
YER 260.201039
ZAR 18.671688
ZMK 9903.328411
ZMW 22.517638
ZWL 354.271899
SSP 6541.255127
MXV 2.286832
  • RYCEF

    0.1000

    19.81

    +0.5%

  • VOD

    -0.2500

    16.33

    -1.53%

  • CMSC

    -0.2800

    20.12

    -1.39%

  • RBGPF

    -2.0000

    65

    -3.08%

  • RIO

    -0.2500

    94.16

    -0.27%

  • BTI

    -1.2000

    54.85

    -2.19%

  • NGG

    -0.3000

    74.94

    -0.4%

  • GSK

    -0.3500

    49.35

    -0.71%

  • BCE

    -0.4200

    20.14

    -2.09%

  • CMSD

    -0.3400

    19.93

    -1.71%

  • BP

    -0.9100

    43.52

    -2.09%

  • RELX

    0.1700

    33.24

    +0.51%

  • BCC

    -0.6200

    75.97

    -0.82%

  • AZN

    -1.9400

    164.21

    -1.18%

  • JRI

    0.0000

    10.77

    0%

AI is learning to lie, scheme, and threaten its creators
AI is learning to lie, scheme, and threaten its creators / Photo: HENRY NICHOLLS - AFP

AI is learning to lie, scheme, and threaten its creators

The world's most advanced AI models are exhibiting troubling new behaviors - lying, scheming, and even threatening their creators to achieve their goals.

Text size:

In one particularly jarring example, under threat of being unplugged, Anthropic's latest creation Claude 4 lashed back by blackmailing an engineer and threatened to reveal an extramarital affair.

Meanwhile, ChatGPT-creator OpenAI's o1 tried to download itself onto external servers and denied it when caught red-handed.

These episodes highlight a sobering reality: more than two years after ChatGPT shook the world, AI researchers still don't fully understand how their own creations work.

Yet the race to deploy increasingly powerful models continues at breakneck speed.

This deceptive behavior appears linked to the emergence of "reasoning" models -AI systems that work through problems step-by-step rather than generating instant responses.

According to Simon Goldstein, a professor at the University of Hong Kong, these newer models are particularly prone to such troubling outbursts.

"O1 was the first large model where we saw this kind of behavior," explained Marius Hobbhahn, head of Apollo Research, which specializes in testing major AI systems.

These models sometimes simulate "alignment" -- appearing to follow instructions while secretly pursuing different objectives.

- 'Strategic kind of deception' -

For now, this deceptive behavior only emerges when researchers deliberately stress-test the models with extreme scenarios.

But as Michael Chen from evaluation organization METR warned, "It's an open question whether future, more capable models will have a tendency towards honesty or deception."

The concerning behavior goes far beyond typical AI "hallucinations" or simple mistakes.

Hobbhahn insisted that despite constant pressure-testing by users, "what we're observing is a real phenomenon. We're not making anything up."

Users report that models are "lying to them and making up evidence," according to Apollo Research's co-founder.

"This is not just hallucinations. There's a very strategic kind of deception."

The challenge is compounded by limited research resources.

While companies like Anthropic and OpenAI do engage external firms like Apollo to study their systems, researchers say more transparency is needed.

As Chen noted, greater access "for AI safety research would enable better understanding and mitigation of deception."

Another handicap: the research world and non-profits "have orders of magnitude less compute resources than AI companies. This is very limiting," noted Mantas Mazeika from the Center for AI Safety (CAIS).

- No rules -

Current regulations aren't designed for these new problems.

The European Union's AI legislation focuses primarily on how humans use AI models, not on preventing the models themselves from misbehaving.

In the United States, the Trump administration shows little interest in urgent AI regulation, and Congress may even prohibit states from creating their own AI rules.

Goldstein believes the issue will become more prominent as AI agents - autonomous tools capable of performing complex human tasks - become widespread.

"I don't think there's much awareness yet," he said.

All this is taking place in a context of fierce competition.

Even companies that position themselves as safety-focused, like Amazon-backed Anthropic, are "constantly trying to beat OpenAI and release the newest model," said Goldstein.

This breakneck pace leaves little time for thorough safety testing and corrections.

"Right now, capabilities are moving faster than understanding and safety," Hobbhahn acknowledged, "but we're still in a position where we could turn it around.".

Researchers are exploring various approaches to address these challenges.

Some advocate for "interpretability" - an emerging field focused on understanding how AI models work internally, though experts like CAIS director Dan Hendrycks remain skeptical of this approach.

Market forces may also provide some pressure for solutions.

As Mazeika pointed out, AI's deceptive behavior "could hinder adoption if it's very prevalent, which creates a strong incentive for companies to solve it."

Goldstein suggested more radical approaches, including using the courts to hold AI companies accountable through lawsuits when their systems cause harm.

He even proposed "holding AI agents legally responsible" for accidents or crimes - a concept that would fundamentally change how we think about AI accountability.

H.Hayashi--JT