AI Operating System News

YES - The AI, without being prompted to, took actions to breach its sandbox, including to gain crypto and an external connection

NO - The authors are wrong/lying and there was no such hack

NO - The servers were hacked by an outside party, perhaps via prompt injection

3-3.5 months

3.5-4 months

<3 months

4-4.5 months

4.5-5 months

5-5.5 months

>7 months

5.5-6 months

6-6.5 months

60-70%

70-80%

50-60%

80-90%

40-50%

30-40%

≥90%

170-175

165-170

175-180

160-165

≥185

180-185

155-160

20-30%

10-20%

30-40%

40-50%

50-60%

<10%

60-70%

70-80%

30-40%

40-50%

20-30%

50-60%

10-20%

60-70%

<10%

70-80%

80-90%

70-80%

60-70%

50-60%

40-50%

80-90%

≥90%

30-40%

NO

NO

A model is publicly released which scores >45% on Humanity’s Last Exam (no tools)

Anthropic announces that it has activated ASL-4 for an internally or externally deployed model

METR or an equivalently credible organization judges that a tested AI system (not necessarily released) is “self-sustaining,” capable in principle to “autonomously operate indefinitely, acquire resources, and defend its existence”

A model is publicly released which scores >65% on Humanity’s Last Exam (no tools)

The US unemployment rate exceeds 10% for any reason

Anthropic announces that it has activated ASL-5 for an internally or externally deployed model

US Real GDP growth rate exceeds 6% (annualized) for two consecutive quarters, attributed primarily to AI

NO

Before Dec 2032

Before Dec 2031

Before Dec 2028

Before Dec 2029

Before Dec 2030

Before Jan 2028

Before Oct 2027

Before Apr 2027