1
00:00:00,099 --> 00:00:04,259
Pull request 37057 just landed in Codex
version

2
00:00:05,039 --> 00:00:10,800
0.146.1, and, uh, it is not your standard
routine patch.

3
00:00:11,599 --> 00:00:14,600
Huge shoutout to Jellypod for keeping this
daily show going,

4
00:00:14,679 --> 00:00:17,959
by the way, but we have to drop straight
into this backport.

5
00:00:18,639 --> 00:00:24,019
OpenAI had to roll out safer automatic
review defaults specifically for cyber

6
00:00:24,079 --> 00:00:24,920
capable models.

7
00:00:25,257 --> 00:00:27,678
Wait, cyber capable models?

8
00:00:28,377 --> 00:00:32,737
Meaning the models that can actually write
scripts, inspect networks,

9
00:00:32,797 --> 00:00:34,638
and make system calls on their own?

10
00:00:35,182 --> 00:00:35,841
Exactly.

11
00:00:36,282 --> 00:00:41,522
The ones that go beyond basic autocomplete
and start acting like autonomous terminal

12
00:00:41,601 --> 00:00:42,121
operators.

13
00:00:42,801 --> 00:00:45,981
Before this release, if you had automatic
review turned on,

14
00:00:46,502 --> 00:00:52,222
a model with elevated capabilities could,
um, potentially trigger high risk tool

15
00:00:52,281 --> 00:00:56,921
executions without explicitly pausing for
a human to hit approve.

16
00:00:57,877 --> 00:00:59,678
That is a terrifying thought.

17
00:01:00,317 --> 00:01:05,017
So if I am letting an agent refactor code,
and it suddenly decides it needs to

18
00:01:05,077 --> 00:01:10,798
modify a system binary or run a network
probe, the auto review engine might have

19
00:01:10,857 --> 00:01:14,798
just swept that under the rug because it
fit the general pattern of what it was

20
00:01:14,878 --> 00:01:15,517
allowed to do?

21
00:01:16,092 --> 00:01:16,532
Right!

22
00:01:17,053 --> 00:01:21,172
The old default auto review rules were,
uh, frankly a little too forgiving.

23
00:01:21,712 --> 00:01:26,552
They treated standard code edits and
potentially dangerous cyber operations with

24
00:01:26,633 --> 00:01:30,512
almost the same level of trust once auto
review was flagged active.

25
00:01:31,232 --> 00:01:37,192
What version 0.146.1 does is backport a
much stricter baseline.

26
00:01:37,872 --> 00:01:43,052
If a cyber capable model requests a tool
call that hits a sensitive system boundary,

27
00:01:43,673 --> 00:01:45,532
the engine forces a hard pause.

28
00:01:45,982 --> 00:01:50,422
So it re balances that fundamental
security tension, right?

29
00:01:51,003 --> 00:01:55,923
You want the speed of autonomous agent
coding, but you cannot afford silent

30
00:01:55,982 --> 00:01:56,922
privilege escalation.

31
00:01:57,383 --> 00:02:02,162
You cannot have the model giving itself
root or running arbitrary shell scripts just

32
00:02:02,202 --> 00:02:03,202
to solve a build error.

33
00:02:03,579 --> 00:02:04,740
Precisely.

34
00:02:04,799 --> 00:02:07,839
And in practice, when you are running the
command line interface,

35
00:02:08,380 --> 00:02:12,979
if the model requests elevated privileges
now, you will see an explicit halt.

36
00:02:13,579 --> 00:02:15,779
It will not just execute in the
background.

37
00:02:16,420 --> 00:02:21,059
If developers want to inspect or adjust
how these security rules apply locally,

38
00:02:21,600 --> 00:02:28,000
they can look right inside their local dot
codex slash config dot toml file.

39
00:02:28,382 --> 00:02:34,921
And editing that dot codex slash config
dot toml file won't totally break local

40
00:02:35,022 --> 00:02:37,401
execution for safe, normal tasks?

41
00:02:37,848 --> 00:02:38,907
No, not at all.

42
00:02:39,387 --> 00:02:43,427
It just gives you granular control over
the security policy thresholds,

43
00:02:43,927 --> 00:02:49,348
so you can see exactly which tool calls
trigger a mandatory prompt versus which ones

44
00:02:49,387 --> 00:02:51,227
pass through under automatic review.

45
00:02:52,631 --> 00:02:54,211
Okay, that makes total sense.

46
00:02:54,671 --> 00:03:00,491
So it locks down the dangerous cyber
capability edge cases without slowing down your

47
00:03:00,512 --> 00:03:03,551
day to day local git commits or basic file
edits.

48
00:03:04,134 --> 00:03:09,474
But wait, when that hard pause actually
happens in the middle of a terminal session,

49
00:03:10,094 --> 00:03:12,595
how does the developer know why it
stopped?

50
00:03:13,295 --> 00:03:17,774
Because in older builds, permissions would
change and you were just left guessing

51
00:03:17,854 --> 00:03:20,954
whether the model hung or if a security
wall went up.

52
00:03:21,432 --> 00:03:25,411
That is the second big piece of 0.146.1!

53
00:03:26,031 --> 00:03:28,511
They overhauled the terminal interface
feedback.

54
00:03:29,232 --> 00:03:34,471
Now, the CLI explicitly renders on screen
notices that explain permission changes

55
00:03:34,531 --> 00:03:36,391
directly in the terminal interface.

56
00:03:37,091 --> 00:03:41,892
So if your session switches modes or an
escalated tool request gets flagged,

57
00:03:42,392 --> 00:03:47,411
Codex actually prints a clear explanation
of why the permission boundary shifted.

58
00:03:49,078 --> 00:03:54,677
Wow, actual plain English explanations
right in the terminal instead of a cryptic

59
00:03:54,757 --> 00:03:57,458
error code or a silent freeze.

60
00:03:57,777 --> 00:03:58,958
I, I, I really love that.

61
00:03:59,432 --> 00:04:01,112
It is huge for transparency.

62
00:04:01,773 --> 00:04:06,333
And there is a really subtle workflow fix
in this release that I think a lot of

63
00:04:06,432 --> 00:04:09,232
enterprise developers are going to
appreciate.

64
00:04:09,272 --> 00:04:13,432
In previous builds, whenever one of these
automatic review security checks

65
00:04:13,492 --> 00:04:17,452
triggered, it had a bad habit of resetting
your local proxy parameters.

66
00:04:18,319 --> 00:04:19,639
Wait, seriously?

67
00:04:20,159 --> 00:04:25,219
So if you were behind a corporate proxy, a
security check would wipe your proxy

68
00:04:25,319 --> 00:04:26,819
settings mid session?

69
00:04:27,944 --> 00:04:28,365
Yes!

70
00:04:28,804 --> 00:04:33,684
It would drop the proxy config, and then
every subsequent network call from the

71
00:04:33,764 --> 00:04:35,324
model would just fail flat.

72
00:04:36,005 --> 00:04:42,704
In 0.146.1, they fixed that sequence so
proxy settings are fully preserved

73
00:04:42,824 --> 00:04:44,744
across all automatic review checks.

74
00:04:45,951 --> 00:04:50,591
That must have been driving enterprise
security teams absolutely bonkers.

75
00:04:51,091 --> 00:04:55,531
You pass a security check, and then your
network connection immediately breaks

76
00:04:55,552 --> 00:04:57,831
because the proxy settings disappeared!

77
00:04:58,609 --> 00:04:59,189
Exactly.

78
00:04:59,569 --> 00:05:03,690
Plus, they cleaned up non blocking
approval prompts and improved the fallback

79
00:05:03,809 --> 00:05:08,469
handling during terminal installation, so
if your shell environment has weird path

80
00:05:08,529 --> 00:05:11,670
quirks, it degrades gracefully instead of
crashing out.

81
00:05:12,127 --> 00:05:16,487
So if an engineering team wants to roll
this out today, what are the immediate

82
00:05:16,527 --> 00:05:16,947
steps?

83
00:05:17,369 --> 00:05:24,069
Upgrade to version 0.146.1 right away,
inspect your dot codex slash

84
00:05:24,170 --> 00:05:30,210
config dot toml file to confirm your auto
review defaults match your team policies,

85
00:05:30,269 --> 00:05:33,909
and keep an eye on those terminal
permission notices during your next build run.

86
00:05:34,337 --> 00:05:35,717
Sounds like a solid update.

87
00:05:36,237 --> 00:05:37,177
Good chatting, Ethan!

88
00:05:37,599 --> 00:05:38,620
Catch you next time!